Offer the full playback rate list on Safari (#24444 )

Rate events over at least one second (#24455 )
* Rate events over at least one second EventsPerSecond.eps() divided the event count by the time since start(), which can be a few milliseconds right after a restart. Frames buffered during an ffmpeg restart then report as 100+ fps, and the same happens to the detector fps. Use a window of at least one second. * Keep sub-second windows consistent Floor the divisor at the window length when the window is shorter than a second, so a caller with a sub-second window still gets its true rate.
2026-09-24 18:26:51 +03:00 · 2026-09-24 10:02:06 -05:00 · 2026-09-24 06:28:00 -06:00 · 2026-09-21 07:50:44 -05:00 · 2026-09-20 12:45:24 -06:00 · 2026-09-19 07:16:00 -05:00
1472 changed files with 104976 additions and 39301 deletions
@@ -8,6 +8,7 @@ amdgpu
 analyzeduration
 Annke
 apexcharts
+Aqara
 arange
 argmax
 argmin
@@ -54,7 +55,6 @@ Dahua
 datasheet
 debconf
 deci
-deepstack
 defragment
 devcontainer
 DEVICEMAP
@@ -64,6 +64,7 @@ dsize
 dtype
 ECONNRESET
 edgetpu
+Eufy
 facenet
 fastapi
 faststart
@@ -26,8 +26,8 @@ body:
    id: version
    attributes:
      label: Beta Version
-      description: Visible on the System Metrics page in the Web UI. Please include the full version including the build identifier (eg. 0.18.0-beta1, 0.18.0-8b72c7a, etc.)
-      placeholder: "0.18.0-beta1"
+      description: Visible on the System Metrics page in the Web UI. Please include the full version including the build identifier (eg. 0.19.0-beta1, 0.19.0-8b72c7a, etc.)
+      placeholder: "0.19.0-beta1"
    validations:
      required: true
  - type: dropdown
@@ -6,7 +6,9 @@ body:
      value: |
        Use this form to submit a reproducible bug in Frigate or Frigate's UI.

-        **⚠️ If you are running a beta version (0.18.0-beta or similar), please use the [Beta Support template](https://github.com/blakeblackshear/frigate/discussions/new?category=beta-support) instead.**
+        If you are running on Proxmox, please see the [Proxmox FAQ](https://github.com/blakeblackshear/frigate/discussions/23916) and reproduce the issue on a standard Docker install first (bare metal, or a VM running plain Debian/Ubuntu) before submitting here.
+
+        **⚠️ If you are running a beta version (0.19.0-beta or similar), please use the [Beta Support template](https://github.com/blakeblackshear/frigate/discussions/new?category=beta-support) instead.**

        Before submitting your bug report, please ask the AI with the "Ask AI" button on the [official documentation site][ai] about your issue, [search the discussions][discussions], look at recent open and closed [pull requests][prs], read the [official Frigate documentation][docs], and read the [Frigate FAQ][faq] pinned at the Discussion page to see if your bug has already been fixed by the developers or reported by the community.

@@ -23,7 +23,7 @@ jobs:
    name: AMD64 Build
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -42,12 +42,391 @@ jobs:
          tags: ${{ steps.setup.outputs.image-name }}-amd64
          cache-from: type=registry,ref=${{ steps.setup.outputs.cache-name }}-amd64
          cache-to: type=registry,ref=${{ steps.setup.outputs.cache-name }}-amd64,mode=max
+  smoke_test:
+    runs-on: ubuntu-22.04
+    name: AMD64 Smoke Test
+    needs:
+      - amd64_build
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v7
+        with:
+          persist-credentials: false
+      - name: Set up QEMU and Buildx
+        id: setup
+        uses: ./.github/actions/setup
+        with:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Start container
+        run: |
+          mkdir -p /tmp/frigate-config /tmp/frigate-media
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config/config.yml
+          # simulate a root-era install: root-owned 0600 jwt secret pre-exists
+          docker run --rm -v /tmp/frigate-config:/config --entrypoint bash \
+            ${{ steps.setup.outputs.image-name }}-amd64 \
+            -c "python3 -c 'import secrets; open(\"/config/.jwt_secret\",\"w\").write(secrets.token_hex(64))' && chmod 600 /config/.jwt_secret && chown 0:0 /config/.jwt_secret"
+          docker run -d --name frigate --shm-size 256m \
+            -v /tmp/frigate-config:/config \
+            -v /tmp/frigate-media:/media/frigate \
+            --mount type=tmpfs,target=/tmp/cache,tmpfs-size=100000000 \
+            -p 5000:5000 -p 8971:8971 \
+            ${{ steps.setup.outputs.image-name }}-amd64
+      - name: Wait for API
+        run: |
+          for i in $(seq 1 60); do
+            curl -fs http://127.0.0.1:5000/api/version && exit 0
+            sleep 5
+          done
+          echo "API never came up"; docker logs frigate; exit 1
+      - name: Assert security headers and permissions
+        run: |
+          headers=$(curl -ksI https://127.0.0.1:8971/)
+          echo "$headers"
+          echo "$headers" | grep -qi "x-content-type-options: nosniff"
+          echo "$headers" | grep -qi "referrer-policy: strict-origin-when-cross-origin"
+          # server_tokens off: Server header must not include a version.
+          # written as an if rather than "! grep", because bash exempts a
+          # negated command from set -e and the assertion would never fail
+          if echo "$headers" | grep -qiE "^server: nginx/[0-9]"; then
+            echo "Server header leaks the nginx version; server_tokens is not off"
+            exit 1
+          fi
+          # Frigate never ships frame-ancestors: HA's Webpage card and iframe
+          # panels frame it cross-origin and it would break them silently
+          if echo "$headers" | grep -qi "frame-ancestors"; then
+            echo "response carries frame-ancestors, which breaks cross-origin iframe embedding"
+            exit 1
+          fi
+          # -t as root would chown the live cache and temp dirs to the `user`
+          # directive user; stdout discarded because -t reopens the config's
+          # /dev/stdout logs and the docker exec pipe is root-owned
+          docker exec frigate /command/s6-setuidgid frigate bash -c '/usr/local/nginx/sbin/nginx -e stderr -t -c /tmp/nginx/conf/nginx.conf >/dev/null'
+          docker exec frigate stat -c %a /config/tls/privkey.pem | grep -qx 600
+          docker exec frigate stat -c %a /dev/shm/go2rtc.yaml | grep -qx 640
+      - name: Assert services run as non-root
+        run: |
+          ps_out=$(docker exec frigate ps -eo user=,comm=)
+          echo "$ps_out"
+          assert_nonroot() {
+            # the process must exist AND no instance of it may run as root
+            echo "$ps_out" | grep -qw "$1" || { echo "$1 is not running"; exit 1; }
+            if echo "$ps_out" | grep -w "$1" | grep -q '^root'; then
+              echo "$1 is running as root"; exit 1
+            fi
+          }
+          assert_nonroot python3
+          assert_nonroot go2rtc
+          assert_nonroot nginx
+          # root-era jwt secret must have been captured by the sweep and the
+          # auth stack must be functional: wrong creds => clean 401, not 500
+          docker exec frigate stat -c %u /config/.jwt_secret | grep -qx "$(docker exec frigate id -u frigate)"
+          code=$(curl -s -o /dev/null -w '%{http_code}' -X POST http://127.0.0.1:5000/api/login \
+            -H 'content-type: application/json' -d '{"user":"admin","password":"definitely-wrong"}')
+          [ "$code" = "401" ] || { echo "login endpoint returned $code"; exit 1; }
+          # a root nginx -t above would have chowned the runtime dirs to root
+          owners=$(docker exec frigate stat -c %U /tmp/nginx /dev/shm/nginx_cache)
+          echo "$owners"
+          if echo "$owners" | grep -qvx frigate; then
+            echo "nginx runtime dirs are not owned by frigate"; exit 1
+          fi
+          # runtime user can write recordings storage
+          docker exec frigate /command/s6-setuidgid frigate touch /media/frigate/.write-probe
+          docker exec frigate rm /media/frigate/.write-probe
+          # tmpfs mount per the docs: arrives root-owned, holds the ZMQ IPC sockets
+          docker exec frigate /command/s6-setuidgid frigate touch /tmp/cache/.write-probe
+          docker exec frigate rm /tmp/cache/.write-probe
+          # models are baked in as root and archive members can carry root-only modes
+          docker exec frigate /command/s6-setuidgid frigate sh -c '
+            for f in /cpu_model.tflite /edgetpu_model.tflite /cpu_audio_model.tflite \
+                     /labelmap.txt /audio-labelmap.txt /openvino-model/*; do
+              [ -e "$f" ] || continue
+              test -r "$f" || { echo "$f is not readable by the runtime user"; exit 1; }
+            done'
+      - name: Assert device access grants
+        run: |
+          # a fake accelerator node created after boot, then the oneshot re-run.
+          # /command is on PATH only for s6-supervised services, and the
+          # with-contenv shebang resolves its execline helpers through PATH
+          docker exec frigate mknod /dev/apex_9 c 120 99
+          docker exec frigate sh -c 'export PATH=/command:$PATH; exec /etc/s6-overlay/s6-rc.d/init-devices/run'
+          acl=$(docker exec frigate getfacl -p /dev/apex_9)
+          echo "$acl"
+          echo "$acl" | grep -q "user:frigate:rw-"
+          echo "$acl" | grep -q "user:go2rtc:rw-"
+          # the usb tree gets recursive grants plus a default ACL that
+          # newly created nodes inherit (the Coral re-enumeration path)
+          docker exec frigate sh -c 'mkdir -p /dev/bus/usb/001 && mknod /dev/bus/usb/001/002 c 189 1'
+          docker exec frigate sh -c 'export PATH=/command:$PATH; exec /etc/s6-overlay/s6-rc.d/init-devices/run'
+          docker exec frigate getfacl -p /dev/bus/usb/001 | grep -q "user:frigate:rwx"
+          docker exec frigate sh -c 'mknod /dev/bus/usb/001/099 c 189 98 && chmod 664 /dev/bus/usb/001/099'
+          inherited=$(docker exec frigate getfacl -p /dev/bus/usb/001/099)
+          echo "$inherited"
+          echo "$inherited" | grep -q "user:frigate:rw-"
+          # getfacl prints granted perms even when the mask clamps them to
+          # nothing, with a trailing "#effective:" comment; a clamped ACL must
+          # fail this assertion, not sneak past it. The check is scoped to the
+          # runtime users because the inherited group:: entry is always clamped
+          # on a non-directory, so an unscoped grep could never pass.
+          if echo "$inherited" | grep -E "^user:(frigate|go2rtc):" | grep -q "effective"; then
+            echo "inherited ACL is mask-clamped and grants no real access"; exit 1
+          fi
+          # hardware that is absent must stay silent: the literal table entries
+          # are not globs, so nullglob does not drop them and only an existence
+          # check keeps them from warning on every boot
+          out=$(docker exec frigate sh -c 'export PATH=/command:$PATH; exec /etc/s6-overlay/s6-rc.d/init-devices/run')
+          echo "$out"
+          if echo "$out" | grep -q "WARN"; then
+            echo "grant warned about device nodes that do not exist"; exit 1
+          fi
+      - name: Assert escape hatch restores root
+        run: |
+          mkdir -p /tmp/frigate-config-root
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-root/config.yml
+          # pre-seed so the absence check proves the rm -f, not a vacuous pass
+          echo "2:1000:1000" > /tmp/frigate-config-root/.permissions_version
+          docker run -d --name frigate-root --shm-size 256m \
+            -e FRIGATE_RUN_AS_ROOT=true \
+            -v /tmp/frigate-config-root:/config \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-root curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "escape hatch container never healthy"; docker logs frigate-root; exit 1; fi
+          ps_out=$(docker exec frigate-root ps -eo user=,comm=)
+          echo "$ps_out"
+          echo "$ps_out" | grep -w python3 | grep -q '^root'
+          echo "$ps_out" | grep -w go2rtc  | grep -q '^root'
+          echo "$ps_out" | grep -w nginx   | grep -q '^root'
+          # an if, not ! test: bash exempts negated commands from set -e
+          if docker exec frigate-root test -f /config/.permissions_version; then
+            echo "escape hatch did not delete the sweep sentinel"; exit 1
+          fi
+          docker rm -f frigate-root
+      - name: Assert granular root services
+        run: |
+          mkdir -p /tmp/frigate-config-granular /tmp/frigate-media-granular
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-granular/config.yml
+          docker run -d --name frigate-granular --shm-size 256m \
+            -e FRIGATE_ROOT_SERVICES=frigate \
+            -v /tmp/frigate-config-granular:/config \
+            -v /tmp/frigate-media-granular:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-granular curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "granular container never became healthy"; docker logs frigate-granular; exit 1; fi
+          ps_out=$(docker exec frigate-granular ps -eo user=,comm=)
+          echo "$ps_out"
+          # the listed service runs as root
+          echo "$ps_out" | grep -w python3 | grep -q '^root'
+          # unlisted services still drop; ifs because set -e exempts negated commands
+          if echo "$ps_out" | grep -w go2rtc | grep -q '^root'; then
+            echo "go2rtc is unexpectedly running as root"; exit 1
+          fi
+          if echo "$ps_out" | grep -w nginx | grep -q '^root'; then
+            echo "nginx is unexpectedly running as root"; exit 1
+          fi
+          # the sweep still ran and the sentinel records the mode
+          docker exec frigate-granular cat /config/.permissions_version | grep -qx "2:1000:1000:frigate"
+          # the root frigate process chowns the db it creates (first-boot immediacy)
+          docker exec frigate-granular stat -c %u /config/frigate.db | grep -qx 1000
+          # plant a root-owned straggler; the per-boot sweep must reclaim it on restart
+          docker exec frigate-granular sh -c 'mkdir -p /media/frigate/clips && touch /media/frigate/clips/straggler.webp'
+          docker restart frigate-granular
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-granular curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "granular container never came back after restart"; docker logs frigate-granular; exit 1; fi
+          docker exec frigate-granular stat -c %u /media/frigate/clips/straggler.webp | grep -qx 1000
+          docker rm -f frigate-granular
+      - name: Assert unknown root service fails fast
+        run: |
+          mkdir -p /tmp/frigate-config-badsvc
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-badsvc/config.yml
+          docker run -d --name frigate-badsvc --shm-size 256m \
+            -e FRIGATE_ROOT_SERVICES=frigatee \
+            -v /tmp/frigate-config-badsvc:/config \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          found=0
+          for i in $(seq 1 12); do
+            if docker logs frigate-badsvc 2>&1 | grep -q "unknown service 'frigatee'"; then found=1; break; fi
+            sleep 5
+          done
+          if [ "$found" -ne 1 ]; then
+            echo "no fail-fast error for an unknown service name"; docker logs frigate-badsvc; exit 1
+          fi
+          # the failed oneshot blocks startup through the dependency chain
+          if docker exec frigate-badsvc curl -fs http://127.0.0.1:5000/api/version; then
+            echo "container came up despite an invalid FRIGATE_ROOT_SERVICES"; exit 1
+          fi
+          docker rm -f frigate-badsvc
+      - name: Assert PUID/PGID remapping
+        run: |
+          mkdir -p /tmp/frigate-config-puid /tmp/frigate-media-puid
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-puid/config.yml
+          docker run -d --name frigate-puid --shm-size 256m \
+            -e PUID=1500 -e PGID=1500 \
+            -v /tmp/frigate-config-puid:/config \
+            -v /tmp/frigate-media-puid:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-puid curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "PUID container never became healthy"; docker logs frigate-puid; exit 1; fi
+          docker exec frigate-puid id -u frigate | grep -qx 1500
+          docker exec frigate-puid id -g frigate | grep -qx 1500
+          docker exec frigate-puid cat /config/.permissions_version | grep -qx "2:1500:1500"
+          # second boot must skip the sweep (sentinel hit). Poll rather than
+          # sleep: the string can only come from the second boot (the first
+          # had no sentinel), so grepping the full log is unambiguous.
+          docker restart frigate-puid
+          ok=0
+          for i in $(seq 1 30); do
+            docker logs frigate-puid 2>&1 | grep -q "already applied" && ok=1 && break
+            sleep 2
+          done
+          if [ "$ok" -ne 1 ]; then echo "sentinel skip never logged"; docker logs frigate-puid; exit 1; fi
+          docker rm -f frigate-puid
+      - name: Assert read-only rootfs with --user works
+        run: |
+          mkdir -p /tmp/frigate-config-ro /tmp/frigate-media-ro
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-ro/config.yml
+          sudo chown -R 1000:1000 /tmp/frigate-config-ro /tmp/frigate-media-ro
+          # /run must allow exec: S6_READ_ONLY_ROOT has s6 copy its service
+          # scripts there and run them, and --tmpfs defaults to noexec
+          docker run -d --name frigate-ro --shm-size 256m \
+            --read-only --tmpfs /tmp:rw,size=1g --tmpfs /run:exec,nosuid,nodev,mode=0755,uid=1000,gid=1000 \
+            --user 1000:1000 \
+            --security-opt no-new-privileges:true \
+            -v /tmp/frigate-config-ro:/config \
+            -v /tmp/frigate-media-ro:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-ro curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "read-only container never healthy"; docker logs frigate-ro; exit 1; fi
+          # an if, not "! grep": bash exempts a negated command from set -e and
+          # the assertion would never fail
+          if docker logs frigate-ro 2>&1 | grep -i "read-only file system"; then
+            echo "a service tried to write to the read-only rootfs"; exit 1
+          fi
+          # the self-signed cert has to land in /config, the only writable path
+          docker exec frigate-ro test -f /config/tls/privkey.pem
+          # and nginx must serve it, which is what proves the templated cert path
+          docker exec frigate-ro curl -ksSI https://127.0.0.1:8971/ >/dev/null
+          # logging must work via the s6-log fallback (no logutil-service as non-root)
+          docker exec frigate-ro test -s /dev/shm/logs/frigate/current
+          # runtime user can write recordings storage
+          docker exec frigate-ro touch /media/frigate/.write-probe
+          docker exec frigate-ro rm /media/frigate/.write-probe
+          docker rm -f frigate-ro
+      - name: Assert PUID with read-only fails fast with clear error
+        run: |
+          docker run -d --name frigate-ro-puid --shm-size 256m \
+            --read-only --tmpfs /tmp:rw,size=1g --tmpfs /run:exec,nosuid,nodev,mode=0755 \
+            -e PUID=1500 -e PGID=1500 \
+            -v /tmp/frigate-config-ro:/config \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          found=0
+          for i in $(seq 1 12); do
+            if docker logs frigate-ro-puid 2>&1 | grep -q "not compatible with read_only"; then found=1; break; fi
+            sleep 5
+          done
+          if [ "$found" -ne 1 ]; then
+            echo "no fail-fast error for PUID with a read-only rootfs"; docker logs frigate-ro-puid; exit 1
+          fi
+          docker rm -f frigate-ro-puid
+      - name: Assert EXTRA_GROUPS with read-only fails fast with clear error
+        run: |
+          docker run -d --name frigate-ro-groups --shm-size 256m \
+            --read-only --tmpfs /tmp:rw,size=1g --tmpfs /run:exec,nosuid,nodev,mode=0755 \
+            -e EXTRA_GROUPS=44 \
+            -v /tmp/frigate-config-ro:/config \
+            -v /tmp/frigate-media-ro:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          found=0
+          for i in $(seq 1 12); do
+            if docker logs frigate-ro-groups 2>&1 | grep -q "EXTRA_GROUPS needs a writable /etc"; then found=1; break; fi
+            sleep 5
+          done
+          if [ "$found" -ne 1 ]; then
+            echo "no fail-fast error for EXTRA_GROUPS with a read-only rootfs"; docker logs frigate-ro-groups; exit 1
+          fi
+          docker rm -f frigate-ro-groups
+      - name: Assert read-only rootfs in the default mode works
+        run: |
+          mkdir -p /tmp/frigate-config-rod /tmp/frigate-media-rod
+          printf 'mqtt:\n  enabled: false\ncameras: {}\n' > /tmp/frigate-config-rod/config.yml
+          docker run -d --name frigate-rod --shm-size 256m \
+            --read-only --tmpfs /tmp:rw,size=1g --tmpfs /run:exec,nosuid,nodev,mode=0755 \
+            --security-opt no-new-privileges:true \
+            -v /tmp/frigate-config-rod:/config \
+            -v /tmp/frigate-media-rod:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-rod curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "read-only default-mode container never healthy"; docker logs frigate-rod; exit 1; fi
+          if docker logs frigate-rod 2>&1 | grep -i "read-only file system"; then
+            echo "a service tried to write to the read-only rootfs"; exit 1
+          fi
+          # the point of this mode over docker's user:: the drop still happens
+          # and go2rtc still gets its own separate user
+          ps_out=$(docker exec frigate-rod ps -eo user=,comm=)
+          echo "$ps_out"
+          for svc in python3 nginx; do
+            if echo "$ps_out" | grep -w "$svc" | grep -q '^root'; then
+              echo "$svc is running as root"; exit 1
+            fi
+          done
+          echo "$ps_out" | grep -w go2rtc | grep -q '^go2rtc'
+          # the ownership sweep still ran and recorded itself in /config
+          docker exec frigate-rod cat /config/.permissions_version | grep -qx "2:1000:1000"
+          # setfacl under a read-only rootfs, which nothing else covers:
+          # init-devices exits early under --user, so that path is never reached
+          docker exec frigate-rod mknod /dev/apex_9 c 120 99
+          docker exec frigate-rod sh -c 'export PATH=/command:$PATH; exec /etc/s6-overlay/s6-rc.d/init-devices/run'
+          docker exec frigate-rod getfacl -p /dev/apex_9 | grep -q "user:frigate:rw-"
+          docker rm -f frigate-rod
+      - name: "Assert switching that install to user: still starts"
+        run: |
+          # the config dir above now holds a go2rtc-owned go2rtc_homekit.yml,
+          # which user: keeps readable but not writable (no supplementary groups)
+          docker run -d --name frigate-rod-user --shm-size 256m \
+            --read-only --tmpfs /tmp:rw,size=1g --tmpfs /run:exec,nosuid,nodev,mode=0755,uid=1000,gid=1000 \
+            --user 1000:1000 \
+            -v /tmp/frigate-config-rod:/config \
+            -v /tmp/frigate-media-rod:/media/frigate \
+            ${{ steps.setup.outputs.image-name }}-amd64
+          up=0
+          for i in $(seq 1 60); do
+            docker exec frigate-rod-user curl -fs http://127.0.0.1:5000/api/version && up=1 && break
+            sleep 5
+          done
+          if [ "$up" -ne 1 ]; then echo "container did not survive the switch to user:"; docker logs frigate-rod-user; exit 1; fi
+          docker logs frigate-rod-user 2>&1 | grep -q "HomeKit pairing changes will not persist"
+          docker rm -f frigate-rod-user
+      - name: Teardown
+        if: always()
+        run: docker rm -f frigate || true
  arm64_build:
    runs-on: ubuntu-22.04-arm
    name: ARM Build
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -82,7 +461,7 @@ jobs:
    name: Jetson Jetpack 6
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -113,7 +492,7 @@ jobs:
      - amd64_build
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -154,7 +533,7 @@ jobs:
      - arm64_build
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -179,7 +558,7 @@ jobs:
      - arm64_build
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up QEMU and Buildx
@@ -211,7 +590,7 @@ jobs:
        with:
          string: ${{ github.repository }}
      - name: Log in to the Container registry
-        uses: docker/login-action@184bdaa0721073962dff0199f1fb9940f07167d1
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
@@ -16,10 +16,10 @@ jobs:
    name: Web - Lint
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
        with:
          persist-credentials: false
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
        with:
          node-version: 20.x
      - run: npm install
@@ -35,10 +35,10 @@ jobs:
    name: Web - Test
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
        with:
          persist-credentials: false
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
        with:
          node-version: 20.x
      - run: npm install
@@ -46,18 +46,15 @@ jobs:
      - name: Build web
        run: npm run build
        working-directory: ./web
-      # - name: Test
-      #   run: npm run test
-      #   working-directory: ./web

  web_e2e:
    name: Web - E2E Tests
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
        with:
          persist-credentials: false
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
        with:
          node-version: 20.x
      - run: npm install
@@ -86,11 +83,11 @@ jobs:
    name: Python Checks
    steps:
      - name: Check out the repository
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Set up Python ${{ env.DEFAULT_PYTHON }}
-        uses: actions/setup-python@v5.4.0
+        uses: actions/setup-python@v7.0.0
        with:
          python-version: ${{ env.DEFAULT_PYTHON }}
      - name: Install requirements
@@ -109,10 +106,10 @@ jobs:
    name: Python Tests
    steps:
      - name: Check out code
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
        with:
          persist-credentials: false
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
        with:
          node-version: 20.x
      - name: Install devcontainer cli
@@ -10,7 +10,7 @@ jobs:
    runs-on: ubuntu-latest

    steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
        with:
          persist-credentials: false
      - id: lowercaseRepo
@@ -18,7 +18,7 @@ jobs:
        with:
          string: ${{ github.repository }}
      - name: Log in to the Container registry
-        uses: docker/login-action@184bdaa0721073962dff0199f1fb9940f07167d1
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
@@ -28,3 +28,7 @@ core
 docs/src/components/DockerComposeGenerator/config/devices.ts
 docs/src/components/DockerComposeGenerator/config/hardware.ts
 docs/src/components/DockerComposeGenerator/config/ports.ts
+
+# GenAI review prompt tester local data (frames from real cameras)
+testing-scripts/genai-review-examples/*
+!testing-scripts/genai-review-examples/README.md
@@ -160,7 +160,7 @@ When reviewing code, do NOT comment on:

 ### Code Quality

- **Linting**: ESLint (see `web/.eslintrc.cjs`)
+- **Linting**: ESLint (see `web/eslint.config.js`)
 - **Formatting**: Prettier with Tailwind CSS plugin
 - **Type Safety**: TypeScript strict mode enabled

@@ -5,3 +5,4 @@
 /docker/rockchip/ @MarcA711
 /docker/rocm/ @harakas
 /docker/hailo8l/ @spanner3003
+/docker/deepx/ @sixfab
@@ -1,7 +1,7 @@
 default_target: local

 COMMIT_HASH := $(shell git log -1 --pretty=format:"%h"|tail -1)
-VERSION = 0.18.0
+VERSION = 0.19.0
 IMAGE_REPO ?= ghcr.io/blakeblackshear/frigate
 GITHUB_REF_NAME ?= $(shell git rev-parse --abbrev-ref HEAD)
 BOARDS= #Initialized empty
@@ -24,7 +24,7 @@ yell
 sigh
 singing
 choir
-sodeling
+yodeling
 chant
 mantra
 child_singing
@@ -0,0 +1,131 @@
+#!/bin/bash
+
+# Installs the DEEPX NPU kernel driver and the DX-RT runtime on the Docker host,
+# then enables the vendor's dxrt.service. A container cannot load kernel
+# modules, so this runs outside the image; the driver creates the /dev/dxrt*
+# nodes and the daemon multiplexes the NPU across host and container.
+#
+# Driver, runtime and firmware versions must agree or inference hangs instead
+# of failing at startup. The set this script installs is pinned in
+# driver_version, runtime_version and firmware_version below; move them
+# together, never one at a time.
+#
+# DEEPX NPU support in Frigate is maintained by Sixfab (https://sixfab.com).
+
+set -euo pipefail
+
+driver_version="v2.6.0"
+# the commit the tag resolves to, since DEEPX signs neither tags nor releases
+# and this is compiled and installed as root. Update both together
+driver_commit="7074748e7104f470b02f517583abba652b3f05fa"
+firmware_version="v2.7.4"
+
+sudo apt-get update
+sudo apt-get install -y git build-essential "linux-headers-$(uname -r)" pciutils wget
+
+if ! lspci -d 1ff4: | grep -q .; then
+    echo "No DEEPX device found on the PCIe bus (lspci -d 1ff4:)."
+    echo "Check that the module is seated correctly before continuing."
+    exit 1
+fi
+
+# fetch the pinned commit rather than cloning the tag, so a retag cannot swap
+# in different source. The build directory is reused so a second run after a
+# failure does not stop on the directory already being there
+mkdir -p dx_rt_npu_linux_driver
+cd dx_rt_npu_linux_driver
+git init -q
+git remote get-url origin > /dev/null 2>&1 ||
+    git remote add origin https://github.com/DEEPX-AI/dx_rt_npu_linux_driver.git
+git fetch --depth 1 origin "${driver_commit}"
+git checkout -q FETCH_HEAD
+
+fetched_commit=$(git rev-parse HEAD)
+if [[ "${fetched_commit}" != "${driver_commit}" ]]; then
+    echo "Fetched commit ${fetched_commit} does not match pinned driver_commit ${driver_commit}."
+    echo "Refusing to build unverified driver source."
+    exit 1
+fi
+
+cd modules
+
+sudo ./build.sh -c install --reload
+
+sudo depmod -A
+
+# dx_dma is the PCIe transport, dxrt_driver the NPU driver on top of it
+for module in dx_dma dxrt_driver; do
+    if ! sudo modprobe "${module}"; then
+        echo "Unable to load the ${module} kernel module, common reasons are:"
+        echo "- Secure Boot is enabled and is rejecting the unsigned module."
+        echo "- The running kernel does not match the installed linux-headers."
+        exit 1
+    fi
+done
+
+if ! compgen -G "/dev/dxrt*" > /dev/null; then
+    echo "Modules loaded but no /dev/dxrt* device node appeared."
+    echo "Run ./sanity_check.sh from the driver repo to diagnose."
+    exit 1
+fi
+
+runtime_version="v3.4.0"
+declare -A runtime_sha256=(
+    [amd64]="736cfef009ce9e974ab1ab610d867239d19d72a426a53e367ddcbd53297b6e20"
+    [arm64]="eb6107f5f02f2ad76ae89f414e8b5f346f34fbc6f0888236136853a26be6f6a0"
+)
+
+runtime_release="${runtime_version#v}"
+deb_arch=$(dpkg --print-architecture)
+deb_file="/tmp/libdxrt-bin_${runtime_release}_${deb_arch}.deb"
+
+wget -qO "${deb_file}" \
+    "https://raw.githubusercontent.com/DEEPX-AI/dx_rt/${runtime_version}/release/${runtime_release}/libdxrt-bin_${runtime_release}_${deb_arch}.deb"
+
+expected_sha256="${runtime_sha256[${deb_arch}]:-}"
+if [[ -z "${expected_sha256}" ]]; then
+    echo "No pinned SHA-256 for architecture ${deb_arch}; refusing to install."
+    exit 1
+fi
+if [[ "$(sha256sum "${deb_file}" | cut -d' ' -f1)" != "${expected_sha256}" ]]; then
+    echo "SHA-256 mismatch for ${deb_file}; refusing to install."
+    exit 1
+fi
+
+sudo dpkg -i "${deb_file}"
+sudo ldconfig
+rm -f "${deb_file}"
+
+sudo cp /usr/share/libdxrt-bin/service/dxrt.service /etc/systemd/system/
+
+# With an endpoint set, dxrtd binds that path only, so the socket goes in a
+# directory Frigate can mount (kept across restarts so the mount stays valid)
+# and a symlink at the default /tmp path keeps host tools that do not set the
+# variable working through their own fallback.
+sudo mkdir -p /etc/systemd/system/dxrt.service.d
+sudo tee /etc/systemd/system/dxrt.service.d/frigate.conf > /dev/null <<'UNIT'
+[Service]
+RuntimeDirectory=dxrt
+RuntimeDirectoryMode=0755
+RuntimeDirectoryPreserve=yes
+Environment=DXRT_DYNAMIC_IPC_ENDPOINT=/run/dxrt/dxrt_dynamic_ipc.sock
+ExecStartPost=/bin/ln -sfn /run/dxrt/dxrt_dynamic_ipc.sock /tmp/dxrt_dynamic_ipc.sock
+UNIT
+
+sudo systemctl daemon-reload
+sudo systemctl enable dxrt.service
+sudo systemctl restart dxrt.service
+
+if ! sudo systemctl is-active --quiet dxrt.service; then
+    echo "dxrt.service did not start. Check: sudo journalctl -u dxrt.service"
+    exit 1
+fi
+
+echo "DEEPX driver and runtime installation complete."
+echo "Driver version:   $(modinfo -F version dxrt_driver) (expected ${driver_version#v})"
+echo "Runtime version:  ${runtime_release}"
+echo "Device node(s):   $(echo /dev/dxrt*)"
+echo
+echo "This driver expects NPU firmware ${firmware_version}. Check it with:"
+echo "  dxrt-cli --status"
+echo "Update the module if it does not match before starting Frigate."
@@ -60,10 +60,10 @@ ARG DEBIAN_FRONTEND
 RUN --mount=type=bind,source=docker/main/build_intel_media_driver.sh,target=/deps/build_intel_media_driver.sh \
    /deps/build_intel_media_driver.sh

-FROM scratch AS go2rtc
+FROM wget AS go2rtc
 ARG TARGETARCH
-WORKDIR /rootfs/usr/local/go2rtc/bin
-ADD --link --chmod=755 "https://github.com/AlexxIT/go2rtc/releases/download/v1.9.14/go2rtc_linux_${TARGETARCH}" go2rtc
+RUN --mount=type=bind,source=docker/main/install_go2rtc.sh,target=/deps/install_go2rtc.sh \
+    /deps/install_go2rtc.sh

 FROM wget AS tempio
 ARG TARGETARCH
@@ -146,6 +146,8 @@ RUN wget -q https://github.com/openvinotoolkit/open_model_zoo/raw/master/data/da
 RUN wget -qO - https://www.kaggle.com/api/v1/models/google/yamnet/tfLite/classification-tflite/1/download | tar xvz && mv 1.tflite cpu_audio_model.tflite
 COPY audio-labelmap.txt .

+RUN chmod -R a+rX /rootfs
+

 FROM wget AS s6-overlay
 ARG TARGETARCH
@@ -200,10 +202,6 @@ RUN pip3 wheel --wheel-dir=/wheels -r /requirements-wheels.txt && \
        pip3 wheel --wheel-dir=/wheels -r /requirements-dev.txt; \
    fi

-# Install HailoRT & Wheels
-RUN --mount=type=bind,source=docker/main/install_hailort.sh,target=/deps/install_hailort.sh \
-    /deps/install_hailort.sh
-
 # Collect deps in a single layer
 FROM scratch AS deps-rootfs
 COPY --from=nginx /usr/local/nginx/ /usr/local/nginx/
@@ -214,7 +212,6 @@ COPY --from=libusb-build /usr/local/lib /usr/local/lib
 COPY --from=tempio /rootfs/ /
 COPY --from=s6-overlay /rootfs/ /
 COPY --from=models /rootfs/ /
-COPY --from=wheels /rootfs/ /
 COPY docker/main/rootfs/ /


@@ -265,6 +262,23 @@ ENV PATH="/usr/local/go2rtc/bin:/usr/local/tempio/bin:/usr/local/nginx/sbin:${PA
 RUN --mount=type=bind,source=docker/main/install_deps.sh,target=/deps/install_deps.sh \
    /deps/install_deps.sh

+# Runtime users. frigate may be remapped at start via PUID/PGID (init-usermod)
+# or replaced entirely with docker's --user. go2rtc is intentionally separate
+# and more restricted. frigate-data is the shared group for /config access.
+# -o tolerates variant base images that already contain uid/gid 1000.
+RUN groupadd -o --gid 1000 frigate \
+    && useradd -o --uid 1000 --gid frigate --no-create-home --shell /usr/sbin/nologin frigate \
+    && groupadd --system go2rtc \
+    && useradd --system --gid go2rtc --no-create-home --shell /usr/sbin/nologin go2rtc \
+    && groupadd --system frigate-data \
+    && usermod -aG frigate-data frigate \
+    && usermod -aG frigate-data go2rtc \
+    && for grp in video render plugdev audio; do \
+         if getent group "$grp" >/dev/null; then \
+           usermod -aG "$grp" frigate && usermod -aG "$grp" go2rtc; \
+         fi; \
+       done
+
 ENV DEFAULT_FFMPEG_VERSION="8.0"
 ENV INCLUDED_FFMPEG_VERSIONS="${DEFAULT_FFMPEG_VERSION}:7.0:5.0"

@@ -275,16 +289,12 @@ RUN wget -q https://bootstrap.pypa.io/get-pip.py -O get-pip.py \
 RUN --mount=type=bind,from=wheels,source=/wheels,target=/deps/wheels \
    pip3 install -U /deps/wheels/*.whl

-# Install Axera Engine
-RUN pip3 install https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3-frigate/axengine-0.1.3-py3-none-any.whl
-
+# The Hailo, MemryX, and Axera runtimes are installed at first start by
+# frigate/util/runtime_deps.py, only when that detector is configured.
+# Axera's native libraries are bind mounted from the host.
 ENV PATH="${PATH}:/usr/bin/axcl"
 ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/usr/lib/axcl"

-# Install MemryX runtime (requires libgomp (OpenMP) in the final docker image)
-RUN --mount=type=bind,source=docker/main/install_memryx.sh,target=/deps/install_memryx.sh \
-    bash -c "bash /deps/install_memryx.sh"
-
 COPY --from=deps-rootfs / /

 RUN ldconfig
@@ -297,6 +307,9 @@ EXPOSE 8555/tcp 8555/udp
 ENV S6_LOGGING_SCRIPT="T 1 n0 s10000000 T"
 # Do not fail on long-running download scripts
 ENV S6_CMD_WAIT_FOR_SERVICES_MAXTIME=0
+# Allow running with a read-only root filesystem: s6 copies its scan dir into
+# /run and executes service scripts from there, so /run must allow exec
+ENV S6_READ_ONLY_ROOT=1

 ENTRYPOINT ["/init"]
 CMD []
@@ -307,6 +320,11 @@ HEALTHCHECK --start-period=300s --start-interval=5s --interval=15s --timeout=5s
 # Frigate deps with Node.js and NPM for devcontainer
 FROM deps AS devcontainer

+# /config here is the developer's bind-mounted checkout, not a data volume, so
+# the prepare ownership sweep must not run: it would chown the source tree to
+# the runtime uid and lock out any container user that isn't 1000.
+ENV FRIGATE_RUN_AS_ROOT=true
+
 # Do not start the actual Frigate service on devcontainer as it will be started by VS Code
 # But start a fake service for simulating the logs
 COPY docker/main/fake_frigate_run /etc/s6-overlay/s6-rc.d/frigate/run
@@ -363,3 +381,7 @@ FROM deps AS frigate

 WORKDIR /opt/frigate/
 COPY --from=rootfs / /
+
+# Pre-compile bytecode so a read-only rootfs doesn't force re-parsing the
+# source tree on every boot (pip-installed packages are already compiled)
+RUN python3 -m compileall -q -j0 /opt/frigate/frigate
@@ -3,7 +3,7 @@
 set -euxo pipefail

 NGINX_VERSION="1.27.4"
-VOD_MODULE_VERSION="1.31"
+VOD_MODULE_VERSION="v1.9.1"
 SECURE_TOKEN_MODULE_VERSION="1.5"
 SET_MISC_MODULE_VERSION="v0.33"
 NGX_DEVEL_KIT_VERSION="v0.3.3"
@@ -31,24 +31,24 @@ wget -nv https://nginx.org/download/nginx-${NGINX_VERSION}.tar.gz
 tar -zxf nginx-${NGINX_VERSION}.tar.gz -C /tmp/nginx --strip-components=1
 rm nginx-${NGINX_VERSION}.tar.gz
 mkdir /tmp/nginx-vod-module
-wget -nv https://github.com/kaltura/nginx-vod-module/archive/refs/tags/${VOD_MODULE_VERSION}.tar.gz
+wget -nv https://github.com/dio-az/nginx-vod-module/archive/refs/tags/${VOD_MODULE_VERSION}.tar.gz
 tar -zxf ${VOD_MODULE_VERSION}.tar.gz -C /tmp/nginx-vod-module --strip-components=1
 rm ${VOD_MODULE_VERSION}.tar.gz
    # Patch MAX_CLIPS to allow more clips to be added than the default 128
 sed -i 's/MAX_CLIPS (128)/MAX_CLIPS (1080)/g' /tmp/nginx-vod-module/vod/media_set.h
 patch -d /tmp/nginx-vod-module/ -p1 << 'EOF'
--- a/vod/avc_hevc_parser.c       2022-06-27 11:38:10.000000000 +0000
-+++ b/vod/avc_hevc_parser.c       2023-01-16 11:25:10.900521298 +0000
-@@ -3,6 +3,9 @@
+--- a/vod/avc_hevc_parser.c
+++ b/vod/avc_hevc_parser.c
+@@ -2,6 +2,9 @@
+
 bool_t
- avc_hevc_parser_rbsp_trailing_bits(bit_reader_state_t* reader)
- {
+ avc_hevc_parser_rbsp_trailing_bits(bit_reader_state_t* reader) {
 +	// https://github.com/blakeblackshear/frigate/issues/4572
 +	return TRUE;
 +
 	uint32_t one_bit;

- 	if (reader->stream.eof_reached)
+ 	if (reader->stream.eof_reached) {
 EOF


@@ -1,10 +1,14 @@
 """Convert the default SSDLite MobileNet v2 model to OpenVINO IR.

 Replaces the legacy openvino-dev Model Optimizer conversion. The TensorFlow
-frontend converts the Object Detection API frozen graph natively; the four TF
-outputs are then repacked into the single [1, 1, 100, 7] DetectionOutput-style
-tensor that Frigate's OpenVINO detector expects, and the input is flipped to
-BGR to match the legacy reverse_input_channels behavior.
+frontend translates the Object Detection API pre and post processors literally,
+producing per-class NonMaxSuppression, NonZero ops and map loops with data
+dependent shapes that the GPU plugin handles very badly. Both are cut out the
+way ssd_v2_support.json used to do it: the preprocessor is an identity at the
+native 300x300 input, and the postprocessor becomes a single fused
+DetectionOutput. The result is the [1, 1, 100, 7] tensor that Frigate's
+OpenVINO detector expects, with the input flipped to BGR to match the legacy
+reverse_input_channels behavior.
 """

 import numpy as np
@@ -12,31 +16,91 @@ import openvino as ov
 from openvino import opset8 as ops
 from openvino.preprocess import PrePostProcessor

+MODEL_DIR = "/models/ssdlite_mobilenet_v2_coco_2018_05_09"
+OUTPUT_PATH = "/models/ssdlite_mobilenet_v2.xml"
+INPUT_SHAPE = [1, 300, 300, 3]
+
+# faster_rcnn_box_coder divides the deltas by pipeline.config's y/x/height/width
+# scales of 10/10/5/5, which DetectionOutput expresses as per-prior variances.
+BOX_VARIANCES = np.float32([0.1, 0.1, 0.2, 0.2])
+
 model = ov.convert_model(
-    "/models/ssdlite_mobilenet_v2_coco_2018_05_09/frozen_inference_graph.pb",
-    input=[("image_tensor:0", [1, 300, 300, 3])],
+    f"{MODEL_DIR}/frozen_inference_graph.pb",
+    input=[("image_tensor:0", INPUT_SHAPE)],
 )

-# rows of (image_id, class_id, score, xmin, ymin, xmax, ymax)
-boxes = model.output("detection_boxes:0").get_node().input_value(0)
-classes = model.output("detection_classes:0").get_node().input_value(0)
-scores = model.output("detection_scores:0").get_node().input_value(0)
+nodes = {op.get_friendly_name(): op for op in model.get_ordered_ops()}
+parameter = model.get_parameters()[0]

-# (ymin,xmin,ymax,xmax) -> (xmin,ymin,xmax,ymax)
-boxes = ops.gather(boxes, [1, 0, 3, 2], 2)
-classes = ops.unsqueeze(classes, 2)
-scores = ops.unsqueeze(scores, 2)
-image_id = ops.multiply(scores, np.float32(0.0))
+preprocessor = nodes["Preprocessor/map/TensorArrayStack/TensorArrayGatherV3"]
+box_deltas = nodes["Postprocessor/Reshape_1"].output(0)
+class_scores = nodes["Postprocessor/convert_scores"].output(0)
+anchors_output = nodes["Postprocessor/Reshape"].output(0)

-detections = ops.concat([image_id, classes, scores, boxes], 2)
-detections = ops.unsqueeze(detections, 1)
+# The anchors only depend on the static input shape, so fold them into a
+# constant and drop the generator subgraph with the rest of the postprocessor.
+probe = ov.Core().compile_model(
+    ov.Model([anchors_output, preprocessor.output(0)], [parameter], "probe"), "CPU"
+)
+probe_input = np.random.default_rng(0).integers(0, 255, INPUT_SHAPE, dtype=np.uint8)
+anchors, resized = (out.copy() for out in probe([probe_input]).values())
+
+assert np.allclose(resized, probe_input, atol=1e-3), (
+    "preprocessor is not an identity at 300x300, it cannot be bypassed"
+)
+
+image = ops.convert(parameter, "f32")
+
+for consumer in list(preprocessor.output(0).get_target_inputs()):
+    consumer.replace_source_output(image.output(0))
+
+# (ymin, xmin, ymax, xmax) -> (xmin, ymin, xmax, ymax)
+priors = anchors[:, [1, 0, 3, 2]].astype(np.float32).reshape(-1)
+variances = np.tile(BOX_VARIANCES, len(anchors))
+proposals = ops.constant(np.stack([priors, variances])[np.newaxis])
+
+# (ty, tx, th, tw) -> (dx, dy, dw, dh) for the CENTER_SIZE decode
+box_logits = ops.reshape(ops.gather(box_deltas, [1, 0, 3, 2], 1), [1, -1], False)
+class_preds = ops.reshape(class_scores, [1, -1], False)
+
+detections = ops.detection_output(
+    box_logits,
+    class_preds,
+    proposals,
+    {
+        "background_label_id": 0,
+        "top_k": 100,
+        "keep_top_k": [100],
+        "nms_threshold": 0.6,
+        "confidence_threshold": 0.3,
+        "code_type": "caffe.PriorBoxParameter.CENTER_SIZE",
+        "share_location": True,
+        "variance_encoded_in_target": False,
+        "normalized": True,
+        "clip_before_nms": False,
+        "clip_after_nms": True,
+        "decrease_label_id": False,
+    },
+)
 detections.output(0).get_tensor().set_names({"detection_out"})

-model = ov.Model([detections], model.get_parameters(), "ssdlite_mobilenet_v2")
+model = ov.Model([detections], [parameter], "ssdlite_mobilenet_v2")

 ppp = PrePostProcessor(model)
 ppp.input().tensor().set_layout(ov.Layout("NHWC"))
 ppp.input().preprocess().reverse_channels()
 model = ppp.build()

-ov.save_model(model, "/models/ssdlite_mobilenet_v2.xml", compress_to_fp16=True)
+# Fail the build rather than silently ship the dynamically shaped graph again.
+op_types = [op.get_type_name() for op in model.get_ordered_ops()]
+assert op_types.count("DetectionOutput") == 1, "postprocessor was not fused"
+
+for dynamic_op in ("NonMaxSuppression", "NonZero", "Loop", "TensorIterator"):
+    assert dynamic_op not in op_types, f"{dynamic_op} left in the graph"
+
+output_shape = model.outputs[0].get_partial_shape()
+assert output_shape.is_static and list(output_shape) == [1, 1, 100, 7], (
+    f"unexpected detector output shape {output_shape}"
+)
+
+ov.save_model(model, OUTPUT_PATH, compress_to_fp16=True)
@@ -10,7 +10,7 @@ apt-get -qq install --no-install-recommends -y \
    gnupg \
    wget \
    lbzip2 \
-    procps vainfo \
+    procps vainfo acl \
    unzip locales tzdata libxml2 xz-utils \
    python3.11 \
    curl \
@@ -28,7 +28,13 @@ update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1
 mkdir -p -m 600 /root/.gnupg

 # install coral runtime
+# sha256 digests of the release debs; update when bumping the libedgetpu release.
+declare -A edgetpu_checksums=(
+    ["amd64"]="63fd00989d29160fa9894e115156a9abe456e88751fc9be89d26e4696200441b"
+    ["arm64"]="eab8aa4576b4dbf738135d8094f32270b24117f77147d25cbe0f49d0144d85f2"
+)
 wget -q -O /tmp/libedgetpu1-max.deb "https://github.com/feranick/libedgetpu/releases/download/16.0TF2.17.1-1/libedgetpu1-max_16.0tf2.17.1-1.bookworm_${TARGETARCH}.deb"
+echo "${edgetpu_checksums[${TARGETARCH}]}  /tmp/libedgetpu1-max.deb" | sha256sum -c -
 unset DEBIAN_FRONTEND
 yes | dpkg -i /tmp/libedgetpu1-max.deb && export DEBIAN_FRONTEND=noninteractive
 rm /tmp/libedgetpu1-max.deb
@@ -45,36 +51,41 @@ if [[ "${TARGETARCH}" == "arm64" ]]; then
    fi
 fi

+# sha256 digests of the ffmpeg builds, keyed "<install dir>-<arch>".
+# Upstream publishes no checksums; these come from a one-time fetch and guard
+# against later substitution. Update when bumping a build URL.
+declare -A ffmpeg_checksums=(
+    ["5.0-amd64"]="377abec133f9d9e8014dee1b91c9684ac8bb0b5b7d80100a57116ff837c4c0d4"
+    ["7.0-amd64"]="e13860eb90409c8218319c928067834ce450128e86f24cfed5cfe91ce6e31037"
+    ["8.0-amd64"]="9bac85054d351cdc89c0a4f45c8ea5c44df94009aabd964b719bbadd56aedae9"
+    ["5.0-arm64"]="57ee475407bad49910ba9b946428396e30cf075ea28a7912fbe1aa2578085af0"
+    ["7.0-arm64"]="16c8b04e9d0ea9c769ad964c4c453fcf05121a1947237329d2e9d8a5e43e2a3c"
+    ["8.0-arm64"]="cd91948468d0f11ce795a2cdaa0c69911bd1db313b49bb19c22512beb88cde69"
+)
+
+# the tarballs nest their binaries under a directory named for the arch, which
+# matches TARGETARCH for both builds we consume
+install_ffmpeg() {
+    local dir="$1" url="$2"
+    mkdir -p "/usr/lib/ffmpeg/${dir}"
+    wget -qO ffmpeg.tar.xz "${url}"
+    echo "${ffmpeg_checksums[${dir}-${TARGETARCH}]}  ffmpeg.tar.xz" | sha256sum -c -
+    tar -xf ffmpeg.tar.xz -C "/usr/lib/ffmpeg/${dir}" --strip-components 1 "${TARGETARCH}/bin/ffmpeg" "${TARGETARCH}/bin/ffprobe"
+    rm -f ffmpeg.tar.xz
+}
+
 # ffmpeg -> amd64
 if [[ "${TARGETARCH}" == "amd64" ]]; then
-    mkdir -p /usr/lib/ffmpeg/5.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2022-07-31-12-37/ffmpeg-n5.1-2-g915ef932a3-linux64-gpl-5.1.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/5.0 --strip-components 1 amd64/bin/ffmpeg amd64/bin/ffprobe
-    rm -rf ffmpeg.tar.xz
-    mkdir -p /usr/lib/ffmpeg/7.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2024-09-19-12-51/ffmpeg-n7.0.2-18-g3e6cec1286-linux64-gpl-7.0.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/7.0 --strip-components 1 amd64/bin/ffmpeg amd64/bin/ffprobe
-    rm -rf ffmpeg.tar.xz
-    mkdir -p /usr/lib/ffmpeg/8.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2026-06-02-14-20/ffmpeg-n8.1.1-9-g58d4114d36-linux64-gpl-8.1.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/8.0 --strip-components 1 amd64/bin/ffmpeg amd64/bin/ffprobe
-    rm -rf ffmpeg.tar.xz
+    install_ffmpeg 5.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2022-07-31-12-37/ffmpeg-n5.1-2-g915ef932a3-linux64-gpl-5.1.tar.xz"
+    install_ffmpeg 7.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2024-09-19-12-51/ffmpeg-n7.0.2-18-g3e6cec1286-linux64-gpl-7.0.tar.xz"
+    install_ffmpeg 8.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2026-06-02-14-20/ffmpeg-n8.1.1-9-g58d4114d36-linux64-gpl-8.1.tar.xz"
 fi

 # ffmpeg -> arm64
 if [[ "${TARGETARCH}" == "arm64" ]]; then
-    mkdir -p /usr/lib/ffmpeg/5.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2022-07-31-12-37/ffmpeg-n5.1-2-g915ef932a3-linuxarm64-gpl-5.1.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/5.0 --strip-components 1 arm64/bin/ffmpeg arm64/bin/ffprobe
-    rm -f ffmpeg.tar.xz
-    mkdir -p /usr/lib/ffmpeg/7.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2024-09-19-12-51/ffmpeg-n7.0.2-18-g3e6cec1286-linuxarm64-gpl-7.0.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/7.0 --strip-components 1 arm64/bin/ffmpeg arm64/bin/ffprobe
-    rm -f ffmpeg.tar.xz
-    mkdir -p /usr/lib/ffmpeg/8.0
-    wget -qO ffmpeg.tar.xz "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2026-06-02-14-20/ffmpeg-n8.1.1-9-g58d4114d36-linuxarm64-gpl-8.1.tar.xz"
-    tar -xf ffmpeg.tar.xz -C /usr/lib/ffmpeg/8.0 --strip-components 1 arm64/bin/ffmpeg arm64/bin/ffprobe
-    rm -f ffmpeg.tar.xz
+    install_ffmpeg 5.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2022-07-31-12-37/ffmpeg-n5.1-2-g915ef932a3-linuxarm64-gpl-5.1.tar.xz"
+    install_ffmpeg 7.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2024-09-19-12-51/ffmpeg-n7.0.2-18-g3e6cec1286-linuxarm64-gpl-7.0.tar.xz"
+    install_ffmpeg 8.0 "https://github.com/NickM-27/FFmpeg-Builds/releases/download/autobuild-2026-06-02-14-20/ffmpeg-n8.1.1-9-g58d4114d36-linuxarm64-gpl-8.1.tar.xz"
 fi

 # arch specific packages
@@ -120,27 +131,56 @@ if [[ "${TARGETARCH}" == "amd64" ]]; then
    apt-get -qq install -y libtbb12

    # install legacy and standard intel compute packages
+    # sha256 digests of the driver debs, taken from the ww<week>.sum asset
+    # compute-runtime ships per release and the checksum.sha256 on npu-driver
+    # v1.19.0; intel-graphics-compiler and level-zero publish none, so those
+    # five are hash-what-you-get. Refresh after a version bump with
+    # `curl -sL <url> | sha256sum`, cross-checking upstream's sum where the
+    # release still has one. npu-driver stopped publishing them after v1.19.0.
+    declare -A intel_checksums=(
+        ["libigdgmm12_22.9.0_amd64.deb"]="9d712f71c18baee076de9961dda71e8089291e1bd0deb5d649ab5ba5de114f97"
+        ["intel-opencl-icd-legacy1_24.35.30872.36_amd64.deb"]="bbe71e4f414259e06a10cde72c29a2bd78d41b2bb2f6f8463b1806797fe66e85"
+        ["intel-level-zero-gpu-legacy1_1.5.30872.36_amd64.deb"]="40dfbd15ab62de036a00824b304a2aa1fa2d81ad60ef83da09cfe3c5a80c429f"
+        ["intel-igc-opencl_1.0.17537.24_amd64.deb"]="dd016400f87fa2b6a9fa9fbcca7eb4a2629174a29de679709f9bec5cede88b0e"
+        ["intel-igc-core_1.0.17537.24_amd64.deb"]="c1e1ecdfe2064c047c552651cfdcdafc504f2033afafba65654338b880048b67"
+        ["intel-opencl-icd_26.14.37833.4-0_amd64.deb"]="2e15eeb4fe9c1bba467a655967373eec6a20dd04cc7159de53c359f17ab53e41"
+        ["libze-intel-gpu1_26.14.37833.4-0_amd64.deb"]="34ce5791160d87ce6d54edb558a4030858ee1dad2afb067b9c5c58d4cde774c6"
+        ["intel-igc-opencl-2_2.32.7+21184_amd64.deb"]="3c9bddbfe558279402bbeaabcf9c63b8de46b956b0ad9625415fd35dda53ad52"
+        ["intel-igc-core-2_2.32.7+21184_amd64.deb"]="64e5230788e3a31e611e8d815a141b1facb91e5f0ef239233ef3f0614bfe3fd6"
+        ["level-zero_1.28.2+u22.04_amd64.deb"]="9015a579abef960166f8e943858d5c81fd4199a960f07260c1da66038257effb"
+        ["intel-driver-compiler-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb"]="8087bfcc0872d7976d0163203c7c783a4176f813c473766587e86c7b34135dff"
+        ["intel-fw-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb"]="740219c03495f8812c03ab74baf8199acf17d13929001105418d4ba226ba2290"
+        ["intel-level-zero-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb"]="f4f5eb97aa7da52c7fec97e4ddfb43aae01703bbadc767bae1f2d4faf342ba42"
+    )
+
+    fetch_intel_deb() {
+        local url="$1" name
+        name=$(basename "$url")
+        wget -q "$url"
+        echo "${intel_checksums[${name}]}  ${name}" | sha256sum -c -
+    }
+
    # see https://github.com/intel/compute-runtime/blob/master/LEGACY_PLATFORMS.md for more info
    # needed core package
-    wget https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/libigdgmm12_22.9.0_amd64.deb
+    fetch_intel_deb https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/libigdgmm12_22.9.0_amd64.deb
    dpkg -i libigdgmm12_22.9.0_amd64.deb
    rm libigdgmm12_22.9.0_amd64.deb

    # legacy compute-runtime packages
-    wget https://github.com/intel/compute-runtime/releases/download/24.35.30872.36/intel-opencl-icd-legacy1_24.35.30872.36_amd64.deb
-    wget https://github.com/intel/compute-runtime/releases/download/24.35.30872.36/intel-level-zero-gpu-legacy1_1.5.30872.36_amd64.deb
-    wget https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.17537.24/intel-igc-opencl_1.0.17537.24_amd64.deb
-    wget https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.17537.24/intel-igc-core_1.0.17537.24_amd64.deb
+    fetch_intel_deb https://github.com/intel/compute-runtime/releases/download/24.35.30872.36/intel-opencl-icd-legacy1_24.35.30872.36_amd64.deb
+    fetch_intel_deb https://github.com/intel/compute-runtime/releases/download/24.35.30872.36/intel-level-zero-gpu-legacy1_1.5.30872.36_amd64.deb
+    fetch_intel_deb https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.17537.24/intel-igc-opencl_1.0.17537.24_amd64.deb
+    fetch_intel_deb https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.17537.24/intel-igc-core_1.0.17537.24_amd64.deb
    # standard compute-runtime packages
-    wget https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/intel-opencl-icd_26.14.37833.4-0_amd64.deb
-    wget https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/libze-intel-gpu1_26.14.37833.4-0_amd64.deb
-    wget https://github.com/intel/intel-graphics-compiler/releases/download/v2.32.7/intel-igc-opencl-2_2.32.7+21184_amd64.deb
-    wget https://github.com/intel/intel-graphics-compiler/releases/download/v2.32.7/intel-igc-core-2_2.32.7+21184_amd64.deb
+    fetch_intel_deb https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/intel-opencl-icd_26.14.37833.4-0_amd64.deb
+    fetch_intel_deb https://github.com/intel/compute-runtime/releases/download/26.14.37833.4/libze-intel-gpu1_26.14.37833.4-0_amd64.deb
+    fetch_intel_deb https://github.com/intel/intel-graphics-compiler/releases/download/v2.32.7/intel-igc-opencl-2_2.32.7+21184_amd64.deb
+    fetch_intel_deb https://github.com/intel/intel-graphics-compiler/releases/download/v2.32.7/intel-igc-core-2_2.32.7+21184_amd64.deb
    # npu packages
-    wget https://github.com/oneapi-src/level-zero/releases/download/v1.28.2/level-zero_1.28.2+u22.04_amd64.deb
-    wget https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-driver-compiler-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb
-    wget https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-fw-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb
-    wget https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-level-zero-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb
+    fetch_intel_deb https://github.com/oneapi-src/level-zero/releases/download/v1.28.2/level-zero_1.28.2+u22.04_amd64.deb
+    fetch_intel_deb https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-driver-compiler-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb
+    fetch_intel_deb https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-fw-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb
+    fetch_intel_deb https://github.com/intel/linux-npu-driver/releases/download/v1.19.0/intel-level-zero-npu_1.19.0.20250707-16111289554_ubuntu22.04_amd64.deb

    dpkg -i *.deb
    rm *.deb
@@ -0,0 +1,19 @@
+#!/bin/bash
+
+set -euxo pipefail
+
+go2rtc_version="1.9.14"
+
+# sha256 digests of the release binaries; update when bumping go2rtc_version.
+declare -A go2rtc_checksums=(
+    ["amd64"]="32d616af226bd731678ffde328b94cfb94e30339bfefc469cfb76323144615a6"
+    ["arm64"]="359fabade8a7a51e81a55fe6df6b0ef81764a5e1d63179577534eaaa71904b50"
+)
+
+dest_dir="/rootfs/usr/local/go2rtc/bin"
+mkdir -p "${dest_dir}"
+
+wget -qO "${dest_dir}/go2rtc" \
+    "https://github.com/AlexxIT/go2rtc/releases/download/v${go2rtc_version}/go2rtc_linux_${TARGETARCH}"
+echo "${go2rtc_checksums[${TARGETARCH}]}  ${dest_dir}/go2rtc" | sha256sum -c -
+chmod 755 "${dest_dir}/go2rtc"
@@ -1,14 +0,0 @@
-#!/bin/bash
-
-set -euxo pipefail
-
-hailo_version="4.21.0"
-
-if [[ "${TARGETARCH}" == "amd64" ]]; then
-    arch="x86_64"
-elif [[ "${TARGETARCH}" == "arm64" ]]; then
-    arch="aarch64"
-fi
-
-wget -qO- "https://github.com/frigate-nvr/hailort/releases/download/v${hailo_version}/hailort-debian12-${TARGETARCH}.tar.gz" | tar -C / -xzf -
-wget -P /wheels/ "https://github.com/frigate-nvr/hailort/releases/download/v${hailo_version}/hailort-${hailo_version}-cp311-cp311-linux_${arch}.whl"
@@ -1,31 +0,0 @@
-#!/bin/bash
-set -e
-
-# Download the MxAccl for Frigate github release
-wget https://github.com/memryx/mx_accl_frigate/archive/refs/tags/v2.1.0.zip -O /tmp/mxaccl.zip
-unzip /tmp/mxaccl.zip -d /tmp
-mv /tmp/mx_accl_frigate-2.1.0 /opt/mx_accl_frigate
-rm /tmp/mxaccl.zip
-
-# Install Python dependencies
-pip3 install -r /opt/mx_accl_frigate/freeze
-
-# Link the Python package dynamically
-SITE_PACKAGES=$(python3 -c "import site; print(site.getsitepackages()[0])")
-ln -s /opt/mx_accl_frigate/memryx "$SITE_PACKAGES/memryx"
-
-# Copy architecture-specific shared libraries
-ARCH=$(uname -m)
-if [[ "$ARCH" == "x86_64" ]]; then
-    cp /opt/mx_accl_frigate/memryx/x86/libmemx.so* /usr/lib/x86_64-linux-gnu/
-    cp /opt/mx_accl_frigate/memryx/x86/libmx_accl.so* /usr/lib/x86_64-linux-gnu/
-elif [[ "$ARCH" == "aarch64" ]]; then
-    cp /opt/mx_accl_frigate/memryx/arm/libmemx.so* /usr/lib/aarch64-linux-gnu/
-    cp /opt/mx_accl_frigate/memryx/arm/libmx_accl.so* /usr/lib/aarch64-linux-gnu/
-else
-    echo "Unsupported architecture: $ARCH"
-    exit 1
-fi
-
-# Refresh linker cache
-ldconfig
@@ -4,6 +4,15 @@ set -euxo pipefail

 s6_version="3.2.1.0"

+# sha256 digests of the release artifacts, from the .sha256 files published at
+# https://github.com/just-containers/s6-overlay/releases/tag/v3.2.1.0
+# Update these when bumping s6_version.
+declare -A s6_checksums=(
+    ["noarch"]="42e038a9a00fc0fef70bf0bc42f625a9c14f8ecdfe77d4ad93281edf717e10c5"
+    ["x86_64"]="8bcbc2cada58426f976b159dcc4e06cbb1454d5f39252b3bb0c778ccf71c9435"
+    ["aarch64"]="c8fd6b1f0380d399422fc986a1e6799f6a287e2cfa24813ad0b6a4fb4fa755cc"
+)
+
 if [[ "${TARGETARCH}" == "amd64" ]]; then
    s6_arch="x86_64"
 elif [[ "${TARGETARCH}" == "arm64" ]]; then
@@ -12,8 +21,15 @@ fi

 mkdir -p /rootfs/

-wget -qO- "https://github.com/just-containers/s6-overlay/releases/download/v${s6_version}/s6-overlay-noarch.tar.xz" |
-    tar -C /rootfs/ -Jxpf -
+download_and_extract() {
+    local arch="$1"
+    local tarball="/tmp/s6-overlay-${arch}.tar.xz"
+    wget -qO "${tarball}" \
+        "https://github.com/just-containers/s6-overlay/releases/download/v${s6_version}/s6-overlay-${arch}.tar.xz"
+    echo "${s6_checksums[${arch}]}  ${tarball}" | sha256sum -c -
+    tar -C /rootfs/ -Jxpf "${tarball}"
+    rm -f "${tarball}"
+}

-wget -qO- "https://github.com/just-containers/s6-overlay/releases/download/v${s6_version}/s6-overlay-${s6_arch}.tar.xz" |
-    tar -C /rootfs/ -Jxpf -
+download_and_extract "noarch"
+download_and_extract "${s6_arch}"
@@ -4,6 +4,14 @@ set -euxo pipefail

 tempio_version="2021.09.0"

+# sha256 digests of the release binaries; update when bumping tempio_version.
+# Upstream publishes no checksums, so these come from a one-time fetch and
+# guard against later substitution rather than the original download.
+declare -A tempio_checksums=(
+    ["amd64"]="b7b93ebfd24c1161cec7aecfad62ab51f2241149358cef354b86cdbc6a60546f"
+    ["aarch64"]="3a5c32981ba68b75ed9b28497429e5a5cecbeb74c3b821b035a48b37609bb895"
+)
+
 if [[ "${TARGETARCH}" == "amd64" ]]; then
    arch="amd64"
 elif [[ "${TARGETARCH}" == "arm64" ]]; then
@@ -13,4 +21,5 @@ fi
 mkdir -p /rootfs/usr/local/tempio/bin

 wget -q -O /rootfs/usr/local/tempio/bin/tempio "https://github.com/home-assistant/tempio/releases/download/${tempio_version}/tempio_${arch}"
+echo "${tempio_checksums[${arch}]}  /rootfs/usr/local/tempio/bin/tempio" | sha256sum -c -
 chmod 755 /rootfs/usr/local/tempio/bin/tempio
@@ -1,4 +1,4 @@
 ruff == 0.15.20

 # types
-types-peewee == 3.17.*
+types-peewee == 4.0.*
@@ -1,17 +1,17 @@
-aiofiles == 24.1.*
-click == 8.1.*
+aiofiles == 25.1.*
+click == 8.5.*
 # FastAPI
 aiohttp == 3.12.*
 starlette == 0.47.*
-starlette-context == 0.4.*
+starlette-context == 0.5.*
 fastapi[standard-no-fastapi-cloud-cli] == 0.116.*
-uvicorn == 0.35.*
+uvicorn == 0.52.*
 slowapi == 0.1.*
-joserfc == 1.2.*
-cryptography == 44.0.*
+joserfc == 1.6.*
+cryptography == 46.0.*
 pathvalidate == 3.3.*
 markupsafe == 3.0.*
-python-multipart == 0.0.26
+python-multipart == 0.0.31
 # Classification Model Training
 tensorflow == 2.19.* ; platform_machine == 'aarch64'
 tensorflow-cpu == 2.19.* ; platform_machine == 'x86_64'
@@ -26,15 +26,15 @@ psutil == 7.1.*
 pydantic == 2.10.*
 git+https://github.com/fbcotter/py3nvml#egg=py3nvml
 pytz == 2025.*
-pyzmq == 26.2.*
+pyzmq == 27.1.*
 ruamel.yaml == 0.18.*
 tzlocal == 5.2
-requests == 2.32.*
+requests == 2.33.*
 types-requests == 2.32.*
 norfair == 2.3.*
 setproctitle == 1.3.*
 ws4py == 0.5.*
-unidecode == 1.3.*
+unidecode == 1.4.*
 titlecase == 2.4.*
 # Image Manipulation
 numpy == 1.26.*
@@ -51,35 +51,26 @@ google-genai == 1.58.*
 ollama == 0.6.*
 openai == 1.65.*
 # push notifications
-py-vapid == 1.9.*
+py-vapid == 1.9.4
 pywebpush == 2.0.*
 # alpr
-pyclipper == 1.3.*
+pyclipper == 1.4.*
 shapely == 2.0.*
 rapidfuzz==3.12.*
-# HailoRT Wheels
-appdirs==1.4.*
+# HailoRT
 argcomplete==2.0.*
 contextlib2==0.6.*
-distlib==0.3.*
-filelock==3.8.*
 future==0.18.*
-importlib-metadata==5.1.*
-importlib-resources==5.1.*
-netaddr==0.8.*
+netaddr==1.3.*
 netifaces==0.10.*
-verboselogs==1.7.*
-virtualenv==20.17.*
-prometheus-client == 0.21.*
+prometheus-client == 0.26.*
 # TFLite
 tflite_runtime @ https://github.com/frigate-nvr/TFlite-builds/releases/download/v2.17.1/tflite_runtime-2.17.1-cp311-cp311-linux_x86_64.whl; platform_machine == 'x86_64'
 tflite_runtime @ https://github.com/feranick/TFlite-builds/releases/download/v2.17.1/tflite_runtime-2.17.1-cp311-cp311-linux_aarch64.whl; platform_machine == 'aarch64'
 # audio transcription
-sherpa-onnx==1.12.*
-faster-whisper==1.1.*
+sherpa-onnx==1.13.*
+faster-whisper==1.2.*
 librosa==0.11.*
 soundfile==0.13.*
-# DeGirum detector
-degirum == 0.16.*
 # Memory profiling
-memray == 1.15.*
+memray == 1.20.*
@@ -1,4 +1,12 @@
 #!/command/with-contenv bash
 # shellcheck shell=bash

-exec logutil-service /dev/shm/logs/certsync
+if [[ "$(id -u)" -eq 0 ]]; then
+    # logutil-service drops to nobody and applies S6_LOGGING_SCRIPT
+    exec logutil-service /dev/shm/logs/certsync
+fi
+
+# Non-root (--user) fallback: logutil-service cannot change UID, so run
+# s6-log directly with the same directives S6_LOGGING_SCRIPT configures.
+# shellcheck disable=SC2086
+exec s6-log ${S6_LOGGING_SCRIPT:-T 1 n0 s10000000 T} /dev/shm/logs/certsync
@@ -6,9 +6,35 @@ set -o errexit -o nounset -o pipefail

 # Logs should be sent to stdout so that s6 can collect them

+# Not `nginx -s reload`: that has root parse /tmp/nginx/conf, which the
+# unprivileged nginx user can rewrite, and nginx chowns path directives on load.
+function reload_nginx() {
+    local pid
+
+    if ! pid=$(cat /tmp/nginx/nginx.pid 2>/dev/null); then
+        echo "[ERROR] No nginx pid file found, not reloading"
+        return 0
+    fi
+
+    if [[ ! "$pid" =~ ^[0-9]+$ ]] || [[ "$(cat "/proc/${pid}/comm" 2>/dev/null)" != "nginx" ]]; then
+        echo "[ERROR] nginx pid file does not name a running nginx process, not reloading"
+        return 0
+    fi
+
+    kill -HUP "$pid"
+}
+
 echo "[INFO] Starting certsync..."

-lefile="/etc/letsencrypt/live/frigate/fullchain.pem"
+# Resolved once, and the condition must stay identical to the nginx run
+# script's. Testing only fullchain.pem here would pick the mounted cert on a
+# half-populated mount that nginx rejected, and the two fingerprints would then
+# never agree, reloading nginx every cycle forever.
+if [ -f /etc/letsencrypt/live/frigate/privkey.pem ] && [ -f /etc/letsencrypt/live/frigate/fullchain.pem ]; then
+    lefile="/etc/letsencrypt/live/frigate/fullchain.pem"
+else
+    lefile="/config/tls/fullchain.pem"
+fi

 tls_enabled=`python3 /usr/local/nginx/get_nginx_settings.py | jq -r .tls.enabled`
 listen_external_port=`python3 /usr/local/nginx/get_nginx_settings.py | jq -r .listen.external_port`
@@ -49,7 +75,7 @@ do
    then
        echo "[INFO] Reloading nginx to refresh TLS certificate"
        echo "$lefile: $leprint"
-        /usr/local/nginx/sbin/nginx -s reload
+        reload_nginx
    fi

    sleep 60
@@ -1,4 +1,12 @@
 #!/command/with-contenv bash
 # shellcheck shell=bash

-exec logutil-service /dev/shm/logs/frigate
+if [[ "$(id -u)" -eq 0 ]]; then
+    # logutil-service drops to nobody and applies S6_LOGGING_SCRIPT
+    exec logutil-service /dev/shm/logs/frigate
+fi
+
+# Non-root (--user) fallback: logutil-service cannot change UID, so run
+# s6-log directly with the same directives S6_LOGGING_SCRIPT configures.
+# shellcheck disable=SC2086
+exec s6-log ${S6_LOGGING_SCRIPT:-T 1 n0 s10000000 T} /dev/shm/logs/frigate
@@ -4,6 +4,24 @@

 set -o errexit -o nounset -o pipefail

+runs_as_root=0
+if [[ "$(id -u)" -eq 0 ]]; then
+    if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]] || /usr/local/bin/service-runs-as-root frigate; then
+        runs_as_root=1
+    fi
+fi
+
+# /root survives s6-setuidgid and breaks cache writes after the drop; set
+# before opt_in_out so the opt-out marker lands where the service will look
+if [[ "$runs_as_root" -eq 0 ]]; then
+    export HOME=/config
+fi
+
+# detector runtimes installed at first start (pip install --user) live under
+# $HOME/.local; the dynamic loader only reads LD_LIBRARY_PATH at exec time
+export LD_LIBRARY_PATH="${LD_LIBRARY_PATH:+${LD_LIBRARY_PATH}:}${HOME}/.local/lib"
+export PATH="${PATH}:${HOME}/.local/bin"
+
 # opt out of openvino telemetry
 if [ -e /usr/local/bin/opt_in_out ]; then
  /usr/local/bin/opt_in_out --opt_out > /dev/null 2>&1
@@ -30,4 +48,8 @@ cd /opt/frigate || echo "[ERROR] Failed to change working directory to /opt/frig

 # Replace the bash process with the Frigate process, redirecting stderr to stdout
 exec 2>&1
-exec python3 -u -m frigate
+if [[ "$(id -u)" -ne 0 || "$runs_as_root" -eq 1 ]]; then
+    exec python3 -u -m frigate
+else
+    exec s6-setuidgid frigate python3 -u -m frigate
+fi
@@ -1,4 +1,12 @@
 #!/command/with-contenv bash
 # shellcheck shell=bash

-exec logutil-service /dev/shm/logs/go2rtc
+if [[ "$(id -u)" -eq 0 ]]; then
+    # logutil-service drops to nobody and applies S6_LOGGING_SCRIPT
+    exec logutil-service /dev/shm/logs/go2rtc
+fi
+
+# Non-root (--user) fallback: logutil-service cannot change UID, so run
+# s6-log directly with the same directives S6_LOGGING_SCRIPT configures.
+# shellcheck disable=SC2086
+exec s6-log ${S6_LOGGING_SCRIPT:-T 1 n0 s10000000 T} /dev/shm/logs/go2rtc
@@ -4,6 +4,20 @@

 set -o errexit -o nounset -o pipefail

+runs_as_root=0
+if [[ "$(id -u)" -eq 0 ]]; then
+    if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]] || /usr/local/bin/service-runs-as-root go2rtc; then
+        runs_as_root=1
+    fi
+fi
+
+# Root via FRIGATE_ROOT_SERVICES only; the escape hatch sweeps nothing and
+# leaves no unprivileged service, so /config/go2rtc stays as safe as pre-drop.
+granular_root=0
+if [[ "$runs_as_root" -eq 1 && "${FRIGATE_RUN_AS_ROOT:-false}" != "true" ]]; then
+    granular_root=1
+fi
+
 # Logs should be sent to stdout so that s6 can collect them

 function get_ip_and_port_from_supervisor() {
@@ -50,42 +64,6 @@ function set_libva_version() {
    export LIBAVFORMAT_VERSION_MAJOR
 }

-function setup_homekit_config() {
-    local config_path="$1"
-
-    if [[ ! -f "${config_path}" ]]; then
-        echo "[INFO] Creating empty config file for HomeKit..."
-        : > "${config_path}"
-    fi
-
-    # Convert YAML to JSON for jq processing
-    local temp_json="/tmp/cache/homekit_config.json"
-    yq eval -o=json "${config_path}" > "${temp_json}" 2>/dev/null || {
-        echo "[WARNING] Failed to convert HomeKit config to JSON, skipping cleanup"
-        return 0
-    }
-
-    # Use jq to extract the homekit section, if it exists
-    local homekit_json
-    homekit_json=$(jq '
-        if has("homekit") then {homekit: .homekit} else null end
-    ' "${temp_json}" 2>/dev/null) || homekit_json="null"
-
-    # If no homekit section, write an empty config file
-    if [[ "${homekit_json}" == "null" ]]; then
-        : > "${config_path}"
-    else
-        # Convert homekit JSON back to YAML and write to the config file
-        echo "${homekit_json}" | yq eval -P - > "${config_path}" 2>/dev/null || {
-            echo "[WARNING] Failed to convert cleaned config to YAML, creating minimal config"
-            : > "${config_path}"
-        }
-    fi
-
-    # Clean up temp files
-    rm -f "${temp_json}"
-}
-
 set_libva_version

 if [[ -f "/dev/shm/go2rtc.yaml" ]]; then
@@ -106,13 +84,23 @@ else
    echo "[WARNING] Unable to remove existing go2rtc config. Changes made to your frigate config file may not be recognized. Please remove the /dev/shm/go2rtc.yaml from your docker host manually."
 fi

-# HomeKit configuration persistence setup
+# HomeKit persistence. The helper is symlink-safe; hand off to go2rtc only when dropping.
 readonly homekit_config_path="/config/go2rtc_homekit.yml"
-setup_homekit_config "${homekit_config_path}"
+if [[ "$(id -u)" -eq 0 && "$runs_as_root" -eq 0 ]]; then
+    python3 /usr/local/go2rtc/prepare_homekit.py "${homekit_config_path}" --chown
+    chown go2rtc:go2rtc /dev/shm/go2rtc.yaml 2>/dev/null || true
+else
+    python3 /usr/local/go2rtc/prepare_homekit.py "${homekit_config_path}"
+fi

 readonly config_path="/config"

-if [[ -x "${config_path}/go2rtc" ]]; then
+# the sweep hands /config to uid 1000, so a root service must not exec from it
+if [[ "$granular_root" -eq 1 && -x "${config_path}/go2rtc" ]]; then
+  echo "[WARN] Ignoring '${config_path}/go2rtc' because FRIGATE_ROOT_SERVICES runs this service as root and /config is owned by the runtime user; using the embedded binary"
+  echo "[WARN] Use FRIGATE_RUN_AS_ROOT=true instead if you need both a custom go2rtc build and root"
+  readonly binary_path="/usr/local/go2rtc/bin/go2rtc"
+elif [[ -x "${config_path}/go2rtc" ]]; then
  readonly binary_path="${config_path}/go2rtc"
  echo "[WARN] Using go2rtc binary from '${binary_path}' instead of the embedded one"
 else
@@ -125,4 +113,8 @@ echo "[INFO] Starting go2rtc..."
 # Use HomeKit config as the primary config so writebacks go there
 # The main config from Frigate will be loaded as a secondary config
 exec 2>&1
-exec "${binary_path}" -config="${homekit_config_path}" -config=/dev/shm/go2rtc.yaml
+if [[ "$(id -u)" -ne 0 || "$runs_as_root" -eq 1 ]]; then
+    exec "${binary_path}" -config="${homekit_config_path}" -config=/dev/shm/go2rtc.yaml
+else
+    exec s6-setuidgid go2rtc "${binary_path}" -config="${homekit_config_path}" -config=/dev/shm/go2rtc.yaml
+fi
@@ -0,0 +1,105 @@
+#!/command/with-contenv bash
+# shellcheck shell=bash
+# Grant the runtime users access to mapped-in device nodes with POSIX ACLs,
+# so --device works without host-side group or udev setup.
+# No-op when: started with --user (euid != 0), FRIGATE_RUN_AS_ROOT=true,
+# or FRIGATE_DEVICE_ACLS=false.
+
+set -o errexit -o nounset -o pipefail
+
+if [[ "$(id -u)" -ne 0 ]]; then
+    exit 0
+fi
+
+if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]]; then
+    exit 0
+fi
+
+if [[ "${FRIGATE_DEVICE_ACLS:-true}" == "false" ]]; then
+    echo "[INFO] FRIGATE_DEVICE_ACLS=false: skipping device access grants"
+    exit 0
+fi
+
+shopt -s nullglob
+
+device_globs=(
+    "/dev/dri/*"
+    "/dev/accel/*"
+    "/dev/apex_*"
+    "/dev/hailo*"
+    "/dev/video*"
+    "/dev/kfd"
+    "/dev/rknpu*"
+    "/dev/mpp_service"
+    "/dev/rga"
+    "/dev/dma_heap/*"
+    "/dev/nvhost*"
+    "/dev/nvmap"
+    "/dev/nvidia*"
+    "/dev/memx*"
+    "/dev/dxrt*"
+)
+
+IFS=',' read -ra extra_globs <<< "${DEVICE_ACL_PATHS:-}"
+for extra in "${extra_globs[@]}"; do
+    extra="${extra//[[:space:]]/}"
+    if [[ -z "$extra" ]]; then
+        continue
+    fi
+    if [[ "$extra" != /dev/* || "$extra" == *..* ]]; then
+        echo "[ERROR] DEVICE_ACL_PATHS entries must be under /dev, got '${extra}'" >&2
+        exit 1
+    fi
+    device_globs+=("$extra")
+done
+
+granted=0
+failed=0
+
+grant() {
+    local node="$1"
+    # nullglob only drops patterns that hold a metacharacter, so a literal
+    # table entry for absent hardware arrives here verbatim. Warn only about
+    # nodes that exist and could not be granted.
+    if [[ ! -e "$node" ]]; then
+        return 0
+    fi
+    local spec="u:frigate:rw,u:go2rtc:rw"
+    # directories need traverse or nothing under them is reachable
+    if [[ -d "$node" ]]; then
+        spec="u:frigate:rwx,u:go2rtc:rwx"
+    fi
+    if setfacl -m "$spec" "$node" 2>/dev/null; then
+        granted=$((granted + 1))
+    else
+        failed=$((failed + 1))
+        echo "[WARN] could not grant device access on ${node}; see EXTRA_GROUPS in the non-root docs for the fallback"
+    fi
+}
+
+for glob in "${device_globs[@]}"; do
+    # shellcheck disable=SC2231
+    for node in $glob; do
+        grant "$node"
+    done
+done
+
+# USB devices re-enumerate (the Coral uploads firmware and reattaches as a new
+# node), so the directories also get a default ACL new nodes inherit. The
+# inherited grant is clamped by the creating mode's group bits, which is rw on
+# udev hosts (0664) and nothing on raw devtmpfs (0600); hardware-verified.
+if [[ -d /dev/bus/usb ]]; then
+    while IFS= read -r -d '' node; do
+        grant "$node"
+    done < <(find /dev/bus/usb -mindepth 1 -print0)
+    while IFS= read -r -d '' dir; do
+        setfacl -d -m "u:frigate:rw,u:go2rtc:rw" "$dir" 2>/dev/null || \
+            echo "[WARN] could not set a default ACL on ${dir}; a re-enumerating USB device may lose access"
+    done < <(find /dev/bus/usb -type d -print0)
+fi
+
+if [[ "$failed" -gt 0 ]]; then
+    echo "[INFO] device access: granted ${granted} node(s), ${failed} failed"
+elif [[ "$granted" -gt 0 ]]; then
+    echo "[INFO] device access: granted ${granted} node(s) to the runtime users"
+fi
@@ -0,0 +1 @@
+oneshot
@@ -0,0 +1 @@
+/etc/s6-overlay/s6-rc.d/init-devices/run
@@ -0,0 +1,104 @@
+#!/command/with-contenv bash
+# shellcheck shell=bash
+# Remap the frigate user to PUID/PGID and register EXTRA_GROUPS.
+# No-op when: started with --user (euid != 0), FRIGATE_RUN_AS_ROOT=true,
+# or PUID/PGID already match. FRIGATE_ROOT_SERVICES is validated here too.
+
+set -o errexit -o nounset -o pipefail
+
+if [[ "$(id -u)" -ne 0 ]]; then
+    # Started with docker --user; the host owns UID mapping entirely.
+    exit 0
+fi
+
+if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]]; then
+    if [[ -n "${FRIGATE_ROOT_SERVICES:-}" ]]; then
+        echo "[INFO] FRIGATE_RUN_AS_ROOT=true: ignoring FRIGATE_ROOT_SERVICES"
+    fi
+    echo "[INFO] FRIGATE_RUN_AS_ROOT=true: skipping user remapping"
+    exit 0
+fi
+
+# a typo must fail the boot, not silently drop a service to non-root
+if [[ -n "${FRIGATE_ROOT_SERVICES:-}" ]]; then
+    IFS=',' read -ra root_services <<< "${FRIGATE_ROOT_SERVICES}"
+    for entry in "${root_services[@]}"; do
+        entry="${entry//[[:space:]]/}"
+        if [[ -z "$entry" ]]; then
+            continue
+        fi
+        case "$entry" in
+            frigate|go2rtc|nginx) ;;
+            *)
+                echo "[ERROR] FRIGATE_ROOT_SERVICES contains unknown service '${entry}'; valid names are frigate, go2rtc, nginx" >&2
+                exit 1
+                ;;
+        esac
+    done
+fi
+
+puid="${PUID:-1000}"
+pgid="${PGID:-1000}"
+
+if ! [[ "$puid" =~ ^[0-9]+$ && "$pgid" =~ ^[0-9]+$ ]]; then
+    echo "[ERROR] PUID and PGID must be numeric, got '${puid}' and '${pgid}'" >&2
+    exit 1
+fi
+
+# Remapping to 0 would make the frigate user root, so every service would keep
+# full privilege while reporting a successful migration.
+if [[ "$puid" -eq 0 || "$pgid" -eq 0 ]]; then
+    echo "[ERROR] PUID/PGID 0 would run the services as root and defeat the privilege separation." >&2
+    echo "[ERROR] Set FRIGATE_RUN_AS_ROOT=true if you want to keep running as root." >&2
+    exit 1
+fi
+
+# Colliding with the go2rtc ids would merge the two users and collapse the
+# separation between the main process and the network-facing restreamer.
+go2rtc_uid="$(id -u go2rtc)"
+go2rtc_gid="$(id -g go2rtc)"
+if [[ "$puid" -eq "$go2rtc_uid" || "$pgid" -eq "$go2rtc_gid" ]]; then
+    echo "[ERROR] PUID/PGID must not equal the go2rtc service ids (${go2rtc_uid}:${go2rtc_gid})." >&2
+    exit 1
+fi
+
+current_uid="$(id -u frigate)"
+current_gid="$(id -g frigate)"
+
+if [[ "$puid" != "$current_uid" || "$pgid" != "$current_gid" ]]; then
+    if [[ ! -w /etc/passwd ]]; then
+        echo "[ERROR] PUID/PGID remapping needs a writable /etc and is not compatible with read_only: true." >&2
+        echo "[ERROR] Either remove read_only and keep PUID, or drop PUID/PGID and use docker's user: ${puid}:${pgid} instead." >&2
+        echo "[ERROR] See https://docs.frigate.video/configuration/non_root for the compatibility matrix." >&2
+        exit 1
+    fi
+    echo "[INFO] Remapping frigate user to ${puid}:${pgid}"
+    groupmod -o -g "$pgid" frigate
+    usermod -o -u "$puid" frigate
+fi
+
+# EXTRA_GROUPS: numeric host GIDs granting device access (e.g. host render/video)
+if [[ -n "${EXTRA_GROUPS:-}" ]]; then
+    # groupadd and usermod -aG both write /etc/group. Checked up front so a
+    # read-only rootfs reports the real problem instead of dying mid-loop.
+    if [[ ! -w /etc/group ]]; then
+        echo "[ERROR] EXTRA_GROUPS needs a writable /etc and is not compatible with read_only: true." >&2
+        echo "[ERROR] Use docker's group_add: with the same GIDs instead; it needs no writes inside the container." >&2
+        echo "[ERROR] See https://docs.frigate.video/configuration/non_root for the compatibility matrix." >&2
+        exit 1
+    fi
+
+    for gid in ${EXTRA_GROUPS//,/ }; do
+        if ! [[ "$gid" =~ ^[0-9]+$ ]] || [[ "$gid" -eq 0 ]]; then
+            echo "[ERROR] EXTRA_GROUPS must be nonzero numeric GIDs, got '${gid}'" >&2
+            exit 1
+        fi
+        if ! getent group "$gid" >/dev/null; then
+            groupadd -o -g "$gid" "frigate-extra-${gid}"
+        fi
+        group_name="$(getent group "$gid" | cut -d: -f1)"
+        usermod -aG "$group_name" frigate
+        usermod -aG "$group_name" go2rtc
+        echo "[INFO] Added frigate and go2rtc to supplementary group ${group_name} (gid ${gid})"
+    done
+fi
@@ -0,0 +1 @@
+oneshot
@@ -0,0 +1 @@
+/etc/s6-overlay/s6-rc.d/init-usermod/run
@@ -7,5 +7,12 @@ set -o errexit -o nounset -o pipefail
 dirs=(/dev/shm/logs/frigate /dev/shm/logs/go2rtc /dev/shm/logs/nginx /dev/shm/logs/certsync)

 mkdir -p "${dirs[@]}"
-chown nobody:nogroup "${dirs[@]}"
+
+# logutil-service drops s6-log to nobody, so the dirs must stay nobody-owned
+# in root mode. Under docker --user we are already the (only) target user,
+# chown would fail, and the plain s6-log fallback in the *-log services
+# writes as us (the mkdir above is sufficient, /dev/shm is 1777).
+if [[ "$(id -u)" -eq 0 ]]; then
+    chown nobody:nogroup "${dirs[@]}"
+fi
 chmod 02755 "${dirs[@]}"
@@ -1,4 +1,12 @@
 #!/command/with-contenv bash
 # shellcheck shell=bash

-exec logutil-service /dev/shm/logs/nginx
+if [[ "$(id -u)" -eq 0 ]]; then
+    # logutil-service drops to nobody and applies S6_LOGGING_SCRIPT
+    exec logutil-service /dev/shm/logs/nginx
+fi
+
+# Non-root (--user) fallback: logutil-service cannot change UID, so run
+# s6-log directly with the same directives S6_LOGGING_SCRIPT configures.
+# shellcheck disable=SC2086
+exec s6-log ${S6_LOGGING_SCRIPT:-T 1 n0 s10000000 T} /dev/shm/logs/nginx
@@ -2,4 +2,4 @@
 set -e

 # Wait for PID file to exist.
-while ! test -f /run/nginx.pid; do sleep 1; done
+while ! test -f /tmp/nginx/nginx.pid; do sleep 1; done
@@ -4,6 +4,13 @@

 set -o errexit -o nounset -o pipefail

+runs_as_root=0
+if [[ "$(id -u)" -eq 0 ]]; then
+    if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]] || /usr/local/bin/service-runs-as-root nginx; then
+        runs_as_root=1
+    fi
+fi
+
 # Logs should be sent to stdout so that s6 can collect them

 echo "[INFO] Starting NGINX..."
@@ -59,38 +66,105 @@ function set_worker_processes() {
        cpus=4
    fi

-    # we need to catch any errors because sed will fail if user has bind mounted a custom nginx file
-    sed -i "s/worker_processes auto;/worker_processes ${cpus};/" /usr/local/nginx/conf/nginx.conf || true
+    sed -i "s/worker_processes auto;/worker_processes ${cpus};/" /tmp/nginx/conf/nginx.conf
 }

+# Rebuilt root-owned every start: a symlink planted by the previously
+# unprivileged nginx would redirect the root cp/tempio writes below onto any
+# root file. rm does not traverse symlinks; the bare mkdir fails closed if raced.
+rm -rf /tmp/nginx
+mkdir /tmp/nginx
+mkdir -p /tmp/nginx/conf /tmp/nginx/client_body /tmp/nginx/proxy \
+         /tmp/nginx/fastcgi /tmp/nginx/uwsgi /tmp/nginx/scgi
+cp -r /usr/local/nginx/conf/. /tmp/nginx/conf/
+
 set_worker_processes

-# ensure the directory for ACME challenges exists
-mkdir -p /etc/letsencrypt/www
-
-# Create self signed certs if needed
+# TLS certs: user-mounted certs at /etc/letsencrypt/live/frigate (documented
+# contract) always win; otherwise fall back to a self-signed cert persisted in
+# /config/tls, which stays writable under a read-only root filesystem.
 letsencrypt_path=/etc/letsencrypt/live/frigate
-mkdir -p $letsencrypt_path
+selfsigned_path=/config/tls

-if [ ! \( -f "$letsencrypt_path/privkey.pem" -a -f "$letsencrypt_path/fullchain.pem" \) ]; then
-    echo "[INFO] No TLS certificate found. Generating a self signed certificate..."
-    openssl req -new -newkey rsa:4096 -days 365 -nodes -x509 \
-        -subj "/O=FRIGATE DEFAULT CERT/CN=*" \
-        -keyout "$letsencrypt_path/privkey.pem" -out "$letsencrypt_path/fullchain.pem" 2>/dev/null
+if [ -f "$letsencrypt_path/privkey.pem" ] && [ -f "$letsencrypt_path/fullchain.pem" ]; then
+    cert_path="$letsencrypt_path"
+else
+    cert_path="$selfsigned_path"
+
+    # Root writing into /config follows any symlink planted there, and /config
+    # is owned by whoever the host mount says, not by root. Generate as the
+    # runtime user wherever we are going to drop to it; the escape hatch keeps
+    # root all the way through, so that path is refused rather than dropped.
+    gen=()
+    if [[ "$(id -u)" -eq 0 && "${FRIGATE_RUN_AS_ROOT:-false}" != "true" ]]; then
+        gen=(s6-setuidgid frigate)
+    elif [[ "$(id -u)" -eq 0 ]]; then
+        for link in "$cert_path" "$cert_path/privkey.pem" "$cert_path/fullchain.pem"; do
+            if [[ -L "$link" ]]; then
+                echo "[ERROR] ${link} is a symlink; refusing to write TLS material through it as root" >&2
+                exit 1
+            fi
+        done
+    fi
+
+    "${gen[@]}" mkdir -p "$cert_path"
+
+    if [ ! \( -f "$cert_path/privkey.pem" -a -f "$cert_path/fullchain.pem" \) ]; then
+        echo "[INFO] No TLS certificate found. Generating a self signed certificate..."
+        "${gen[@]}" openssl req -new -newkey rsa:4096 -days 365 -nodes -x509 \
+            -subj "/O=FRIGATE DEFAULT CERT/CN=*" \
+            -keyout "$cert_path/privkey.pem" -out "$cert_path/fullchain.pem" 2>/dev/null
+        "${gen[@]}" chmod 600 "$cert_path/privkey.pem"
+        "${gen[@]}" chmod 644 "$cert_path/fullchain.pem"
+    fi
 fi

-# build templates for optional FRIGATE_BASE_PATH environment variable
-python3 /usr/local/nginx/get_nginx_settings.py | \
-    tempio -template /usr/local/nginx/templates/base_path.gotmpl \
-           -out /usr/local/nginx/conf/base_path.conf
+# ACME challenges are only served from a writable rootfs; skipping the mkdir
+# under read_only leaves the location 404ing, which is the same as unused
+mkdir -p /etc/letsencrypt/www 2>/dev/null || true

-# build templates for additional network settings
-python3 /usr/local/nginx/get_nginx_settings.py | \
+# nginx settings are read once; both templates consume them
+nginx_settings=$(python3 /usr/local/nginx/get_nginx_settings.py)
+
+# build templates for optional FRIGATE_BASE_PATH environment variable
+echo "$nginx_settings" | \
+    tempio -template /usr/local/nginx/templates/base_path.gotmpl \
+           -out /tmp/nginx/conf/base_path.conf
+
+# build templates for additional network settings; listen.conf is the only
+# template that needs the resolved cert directory
+echo "$nginx_settings" | \
+    jq --arg p "$cert_path" '.tls.cert_path = $p' | \
    tempio -template /usr/local/nginx/templates/listen.gotmpl \
-           -out /usr/local/nginx/conf/listen.conf
+           -out /tmp/nginx/conf/listen.conf
+
+if [[ "$(id -u)" -eq 0 && "$runs_as_root" -eq 0 ]]; then
+    chown -R frigate:frigate /tmp/nginx
+    # heal the cache: a root `nginx -t` chowns every cycle path to the `user` directive user
+    if [ -d /dev/shm/nginx_cache ]; then
+        chown -R frigate:frigate /dev/shm/nginx_cache
+    fi
+    # nginx reopens /dev/stdout by path for its logs, and s6 made the pipe
+    # root-owned 0600; without this the non-root master exits EACCES
+    chown frigate /dev/stdout
+    # Only mounted certs need handing over; the self-signed pair is already
+    # owned by the runtime user that generated it. Never chown the /config copy:
+    # chown follows symlinks, so it would retarget onto any root file the
+    # runtime user pointed it at. Tolerant because mounted certs may be :ro.
+    if [ "$cert_path" = "$letsencrypt_path" ] && [ -f "$cert_path/privkey.pem" ]; then
+        chown frigate:frigate "$cert_path/privkey.pem" "$cert_path/fullchain.pem" 2>/dev/null || true
+    fi
+fi

 # Replace the bash process with the NGINX process, redirecting stderr to stdout
 exec 2>&1
-exec \
-    s6-notifyoncheck -t 30000 -n 1 \
-    nginx
+# -e stderr: the compiled-in error log path is not writable by the runtime user
+if [[ "$(id -u)" -ne 0 || "$runs_as_root" -eq 1 ]]; then
+    exec \
+        s6-notifyoncheck -t 30000 -n 1 \
+        nginx -e stderr -c /tmp/nginx/conf/nginx.conf
+else
+    exec \
+        s6-notifyoncheck -t 30000 -n 1 \
+        s6-setuidgid frigate nginx -e stderr -c /tmp/nginx/conf/nginx.conf
+fi
@@ -144,3 +144,66 @@ rm -f /dev/shm/.frigate-is-stopping

 migrate_addon_config_dir
 migrate_db_from_media_to_config
+
+# Align volume ownership with the runtime user (one sweep per PUID/schema
+# change, guarded by the sentinel; see fix-ownership). The escape hatch
+# deletes the sentinel instead: ownership is never mutated while it is on,
+# so the next non-root boot must re-sweep whatever root created meanwhile.
+if [[ "$(id -u)" -eq 0 ]]; then
+    if [[ "${FRIGATE_RUN_AS_ROOT:-false}" == "true" ]]; then
+        rm -f /config/.permissions_version
+    else
+        # Only when a mount backs /media/frigate itself: under a parent /media
+        # mount, a dedicated volume added later would be shadowed and skipped
+        sentinel_args=(--sentinel /config/.permissions_version)
+        root_services_mode=""
+        if [[ -n "${FRIGATE_ROOT_SERVICES:-}" ]]; then
+            # || true: an all-empty list (",") fails grep -v and errexit would kill the boot
+            root_services_mode=$(tr ',' '\n' <<< "${FRIGATE_ROOT_SERVICES//[[:space:]]/}" | grep -v '^$' | sort -u | paste -sd, - || true)
+            if [[ -n "$root_services_mode" ]]; then
+                sentinel_args+=(--mode "$root_services_mode")
+            fi
+        fi
+        if ! awk '$2 == "/media/frigate" || $2 ~ /^\/media\/frigate\//' /proc/mounts | grep -q .; then
+            sentinel_args=()
+        fi
+        /usr/local/bin/fix-ownership "${sentinel_args[@]}" \
+            "${PUID:-1000}" "${PGID:-1000}" /config /media/frigate
+
+        # Root services write clips stragglers and caches mid-run; realign the
+        # small trees every boot. Recordings are chowned at create instead.
+        if [[ -n "$root_services_mode" ]]; then
+            # only sweep what exists; clips and exports appear after the first run
+            boot_sweep_paths=(/config)
+            for extra in /media/frigate/clips /media/frigate/exports; do
+                if [[ -d "$extra" ]]; then
+                    boot_sweep_paths+=("$extra")
+                fi
+            done
+            /usr/local/bin/fix-ownership \
+                "${PUID:-1000}" "${PGID:-1000}" "${boot_sweep_paths[@]}"
+        fi
+    fi
+fi
+
+# Must stay after the sweep, which reads an absent /media/frigate as an
+# unmounted volume rather than a swept one
+if [[ "$(id -u)" -eq 0 && ! -d /media/frigate ]]; then
+    # The image does not ship this directory, so on a read-only rootfs it can
+    # only come from a mount. Report that rather than failing under errexit.
+    if ! mkdir -p /media/frigate 2>/dev/null; then
+        echo "[ERROR] /media/frigate does not exist and could not be created, which is what happens with read_only: true and no recordings volume." >&2
+        echo "[ERROR] Mount a volume at /media/frigate." >&2
+        echo "[ERROR] See https://docs.frigate.video/configuration/non_root for the compatibility matrix." >&2
+        exit 1
+    fi
+    if [[ "${FRIGATE_RUN_AS_ROOT:-false}" != "true" ]]; then
+        chown "${PUID:-1000}:${PGID:-1000}" /media/frigate
+    fi
+fi
+
+# usually a tmpfs mount: root-owned on arrival and outside the swept volumes
+if [[ "$(id -u)" -eq 0 && "${FRIGATE_RUN_AS_ROOT:-false}" != "true" ]]; then
+    mkdir -p /tmp/cache
+    chown "${PUID:-1000}:${PGID:-1000}" /tmp/cache
+fi
@@ -0,0 +1,194 @@
+#!/bin/bash
+# Single source of truth for aligning volume ownership with the runtime user.
+#
+# Usage: fix-ownership [--dry-run] [--sentinel FILE] [--mode STRING] UID GID PATH [PATH...]
+#
+#   --dry-run    report what would change, touch nothing
+#   --sentinel   skip entirely when FILE already records "SCHEMA:UID:GID";
+#                write it after a successful run (used by the boot path so
+#                multi-TB volumes are swept once per UID/schema change, not
+#                on every boot)
+#   --mode       append STRING to the sentinel, so changing it re-sweeps once
+#
+# Only files whose uid OR gid differs are touched, so re-runs are cheap.
+# lost+found is skipped: fsck fills it with root-only recovered fragments.
+# Top-level /config additionally grants group frigate-data TRAVERSE ONLY
+# (g+rx) so the separate go2rtc user can reach its pre-created HomeKit file
+# on hosts where /config is mounted 0700. Never g+w: directory write means
+# unlink rights over frigate.db/config.yml, and would let a compromised
+# go2rtc plant /config/go2rtc, which the go2rtc run script executes
+# preferentially, as root under the escape hatch.
+
+set -o errexit -o nounset -o pipefail
+
+# Permissions-layout epoch. Bump to force a one-time re-sweep on upgrade
+# (e.g. when the privilege-drop release must capture files created as root
+# since the previous sweep).
+schema=2
+
+dry_run=0
+sentinel=""
+mode=""
+
+while [[ "${1:-}" == --* ]]; do
+    case "$1" in
+        --dry-run) dry_run=1; shift ;;
+        --sentinel)
+            if [[ -z "${2:-}" ]]; then
+                echo "[ERROR] fix-ownership: --sentinel requires a file argument" >&2
+                exit 2
+            fi
+            sentinel="$2"; shift 2 ;;
+        --mode)
+            if [[ -z "${2:-}" ]]; then
+                echo "[ERROR] fix-ownership: --mode requires a value" >&2
+                exit 2
+            fi
+            mode="$2"; shift 2 ;;
+        *) echo "[ERROR] fix-ownership: unknown option $1" >&2; exit 2 ;;
+    esac
+done
+
+if [[ $# -lt 3 ]]; then
+    echo "Usage: fix-ownership [--dry-run] [--sentinel FILE] [--mode STRING] UID GID PATH..." >&2
+    exit 2
+fi
+
+target_uid="$1"
+target_gid="$2"
+shift 2
+
+if [[ "$(id -u)" -ne 0 ]]; then
+    echo "[INFO] fix-ownership: not running as root, skipping (ownership is managed by the host in --user mode)"
+    exit 0
+fi
+
+# The list folds into the sentinel so entering or leaving a granular root mode
+# re-sweeps once, catching whatever the other ownership mechanisms missed.
+sentinel_content="${schema}:${target_uid}:${target_gid}"
+if [[ -n "$mode" ]]; then
+    sentinel_content="${sentinel_content}:${mode}"
+fi
+
+# safe-sentinel reports only a root-owned regular file, so a forged or
+# symlinked sentinel in the runtime-user-owned /config can't suppress the sweep
+if [[ "$dry_run" -eq 0 && -n "$sentinel" ]]; then
+    if existing=$(/usr/local/bin/safe-sentinel read "$sentinel" 2>/dev/null) && \
+        [[ "$existing" == "$sentinel_content" ]]; then
+        echo "[INFO] fix-ownership: ${target_uid}:${target_gid} (schema ${schema}) already applied, skipping"
+        exit 0
+    fi
+fi
+
+# A sweep that could not chown everything must not be recorded as complete:
+# the sentinel would make every later boot skip it and the entries would stay
+# unreachable once services run unprivileged.
+swept_clean=1
+
+# Entries another mechanism deliberately owns. Chowning them undoes that work
+# and leaves the same "mismatch" waiting for the next boot, so /config could
+# never report itself clean: /config is chgrp'd to frigate-data below so go2rtc
+# can traverse it, and the HomeKit file is handed to the go2rtc user by the
+# go2rtc service. Only the GROUP on /config is exempt; a root-owned /config
+# must still be chowned or the runtime user cannot write there at all.
+# Shared by the counting and the chowning walk so the two cannot disagree.
+mismatch_expr=(
+    "(" -not -uid "$target_uid"
+        -o "(" -not -gid "$target_gid" -a ! -path /config ")"
+    ")"
+    -a ! -path /config/go2rtc_homekit.yml
+)
+if [[ -n "$sentinel" ]]; then
+    # safe-sentinel keeps the sentinel root-owned on purpose and rejects one
+    # owned by anybody else, so chowning it here would suppress the skip and
+    # make every boot re-sweep. Only the trailing write puts it back today.
+    mismatch_expr+=(-a ! -path "$sentinel")
+fi
+
+for path in "$@"; do
+    # An absent root is an incomplete sweep, not a finished one: /media/frigate
+    # is not in the image, so a boot before the volume is mounted would
+    # otherwise record success and the volume would never be swept once added.
+    if [[ ! -d "$path" ]]; then
+        swept_clean=0
+        echo "[WARN] fix-ownership: $path does not exist, skipping; will retry on next boot"
+        continue
+    fi
+
+    echo "[INFO] fix-ownership: scanning ${path} for ownership mismatches; this may take a while on large filesystems"
+
+    # find may fail mid-walk on a live volume (file deleted under it) or on a
+    # stale mount. Tolerate it rather than aborting under errexit, but never
+    # read a failed scan as "nothing to do": that would record the sweep as
+    # complete without having looked.
+    if ! count=$(find "$path" -name lost+found -prune -o "${mismatch_expr[@]}" -printf '.' 2>/dev/null | wc -c); then
+        swept_clean=0
+        echo "[WARN] fix-ownership: could not scan ${path}; will retry on next boot"
+        continue
+    fi
+
+    if [[ "$count" -eq 0 ]]; then
+        echo "[INFO] fix-ownership: $path already owned by ${target_uid}:${target_gid}, nothing to do"
+        continue
+    fi
+
+    # find does not descend symlinks and chown -h retargets the link itself, so
+    # anything behind a symlinked directory is outside this sweep. Following
+    # them is not an option: a link could walk the chown out of the volume.
+    if [[ -n "$(find "$path" -type l -xtype d -print -quit 2>/dev/null)" ]]; then
+        echo "[WARN] fix-ownership: ${path} contains symlinked directories; ownership behind them is not managed and must be aligned by hand"
+    fi
+
+    echo "[WARN] fix-ownership: adjusting ownership of ${count} entries under ${path}"
+    if [[ "$dry_run" -eq 1 ]]; then
+        echo "[INFO] fix-ownership: dry run, not changing ${path}"
+        continue
+    fi
+
+    # -execdir chowns from the entry's own directory, so a parent swapped for a
+    # symlink mid-walk can't redirect the chown out of the volume
+    started=$SECONDS
+    if find "$path" -name lost+found -prune -o "${mismatch_expr[@]}" \
+        -print -execdir chown -h "${target_uid}:${target_gid}" {} + \
+        | awk -v total="$count" -v path="$path" '
+            BEGIN { next_pct = 5 }
+            {
+                pct = int(NR * 100 / total)
+                if (pct > 100) pct = 100
+                if (pct >= next_pct) {
+                    printf "[INFO] fix-ownership: %s %d%% (%d/%d entries)\n", path, pct, NR, total
+                    # mawk block-buffers to a pipe; without fflush the whole
+                    # progress log arrives at once
+                    fflush()
+                    while (next_pct <= pct) next_pct += 5
+                }
+            }'; then
+        elapsed=$((SECONDS - started))
+        if [[ "$elapsed" -ge 60 ]]; then
+            elapsed="$((elapsed / 60))m $((elapsed % 60))s"
+        else
+            elapsed="${elapsed}s"
+        fi
+        echo "[INFO] fix-ownership: finished ${path} in ${elapsed}"
+    else
+        swept_clean=0
+        echo "[WARN] fix-ownership: some entries under ${path} could not be updated (deleted mid-sweep or chown denied); will retry on next mismatch"
+    fi
+done
+
+# go2rtc (separate user) must be able to REACH its HomeKit state in /config.
+# Write access is per-file, not per-directory: go2rtc's PatchConfig rewrites
+# the first -config file via os.WriteFile (in-place truncate, no rename,
+# verified against go2rtc v1.9.14 internal/app/config.go), and the file is
+# always pre-created by setup_homekit_config before go2rtc starts, so
+# O_CREATE never needs directory write. See header comment for why g+w is
+# forbidden here.
+if [[ "$dry_run" -eq 0 && -d /config ]]; then
+    chgrp frigate-data /config 2>/dev/null || true
+    chmod g+rx /config 2>/dev/null || true
+fi
+
+if [[ "$dry_run" -eq 0 && -n "$sentinel" && "$swept_clean" -eq 1 ]]; then
+    /usr/local/bin/safe-sentinel write "$sentinel" "$sentinel_content" || \
+        echo "[WARN] fix-ownership: could not write ${sentinel}; the sweep will run again on next boot"
+fi
@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+"""Read or write the ownership sweep sentinel without following symlinks.
+
+The sentinel lives in /config, which the unprivileged runtime user owns, so it
+can be swapped for a symlink. read trusts only a root-owned regular file; write
+never follows a symlink or fifo onto another file.
+
+Usage:
+  safe-sentinel read PATH            print content, exit 0 only if root-owned regular file
+  safe-sentinel write PATH CONTENT   write CONTENT to a regular file at PATH
+"""
+
+import errno
+import os
+import stat
+import sys
+
+MODE = 0o644
+
+
+def do_read(path: str) -> int:
+    try:
+        fd = os.open(path, os.O_RDONLY | os.O_NOFOLLOW)
+    except OSError:
+        return 1
+    try:
+        st = os.fstat(fd)
+        if not stat.S_ISREG(st.st_mode) or st.st_uid != 0:
+            return 1
+        sys.stdout.buffer.write(os.read(fd, 4096))
+    finally:
+        os.close(fd)
+    return 0
+
+
+def do_write(path: str, content: str) -> int:
+    # O_NONBLOCK so a fifo fails fast (ENXIO) instead of blocking the open.
+    flags = os.O_WRONLY | os.O_CREAT | os.O_NOFOLLOW | os.O_NONBLOCK
+    replace = (errno.ELOOP, errno.ENXIO)
+    try:
+        fd = os.open(path, flags, MODE)
+        if not stat.S_ISREG(os.fstat(fd).st_mode):
+            os.close(fd)
+            raise OSError(errno.ELOOP, "not a regular file")
+    except OSError as err:
+        if err.errno not in replace:
+            raise
+        os.unlink(path)
+        fd = os.open(path, flags | os.O_EXCL, MODE)
+    try:
+        os.ftruncate(fd, 0)
+        os.write(fd, content.encode())
+        # keep it root-owned so a later sweep that chowned the old sentinel to
+        # the runtime user can't make the next read reject and re-sweep
+        os.fchown(fd, 0, 0)
+    finally:
+        os.close(fd)
+    return 0
+
+
+def main(argv: list[str]) -> int:
+    if len(argv) == 3 and argv[1] == "read":
+        return do_read(argv[2])
+    if len(argv) == 4 and argv[1] == "write":
+        try:
+            return do_write(argv[2], argv[3])
+        except OSError:
+            return 1
+    print("usage: safe-sentinel read PATH | write PATH CONTENT", file=sys.stderr)
+    return 2
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
@@ -0,0 +1,18 @@
+#!/bin/bash
+# Exit 0 when FRIGATE_ROOT_SERVICES names the given service. Membership only:
+# the euid and FRIGATE_RUN_AS_ROOT checks stay in the callers.
+#
+# Usage: service-runs-as-root SERVICE
+
+set -o nounset
+
+service="${1:?usage: service-runs-as-root SERVICE}"
+
+IFS=',' read -ra entries <<< "${FRIGATE_ROOT_SERVICES:-}"
+for entry in "${entries[@]}"; do
+    entry="${entry//[[:space:]]/}"
+    if [[ "$entry" == "$service" ]]; then
+        exit 0
+    fi
+done
+exit 1
@@ -3,13 +3,12 @@
 import json
 import os
 import sys
-from pathlib import Path
 from typing import Any

 from ruamel.yaml import YAML

 sys.path.insert(0, "/opt/frigate")
-from frigate.config.env import substitute_frigate_vars
+from frigate.config.env import apply_config_env_vars, substitute_frigate_vars
 from frigate.const import (
    BIRDSEYE_PIPE,
    LIBAVFORMAT_VERSION_MAJOR,
@@ -25,15 +24,6 @@ sys.path.remove("/opt/frigate")

 yaml = YAML()

-FRIGATE_ENV_VARS = {k: v for k, v in os.environ.items() if k.startswith("FRIGATE_")}
-# read docker secret files as env vars too
-if os.path.isdir("/run/secrets"):
-    for secret_file in os.listdir("/run/secrets"):
-        if secret_file.startswith("FRIGATE_"):
-            FRIGATE_ENV_VARS[secret_file] = (
-                Path(os.path.join("/run/secrets", secret_file)).read_text().strip()
-            )
-
 config_file = find_config_file()

 try:
@@ -47,6 +37,20 @@ try:
 except FileNotFoundError:
    config: dict[str, Any] = {}

+# No validator runs here, so install environment_vars ourselves. FRIGATE_
+# names only: anything else lands in os.environ, where the exec gate reads
+# GO2RTC_ALLOW_ARBITRARY_EXEC.
+config_env_vars = config.get("environment_vars")
+apply_config_env_vars(
+    {
+        key: value
+        for key, value in config_env_vars.items()
+        if str(key).startswith("FRIGATE_")
+    }
+    if isinstance(config_env_vars, dict)
+    else {}
+)
+
 go2rtc_config: dict[str, Any] = config.get("go2rtc", {})

 # Need to enable CORS for go2rtc so the frigate integration / card work automatically
@@ -113,7 +117,7 @@ for name in list(go2rtc_config.get("streams", {})):

    if isinstance(stream, str):
        try:
-            formatted_stream = stream.format(**FRIGATE_ENV_VARS)
+            formatted_stream = substitute_frigate_vars(stream)
            if is_restricted_go2rtc_source(formatted_stream):
                print(
                    f"[ERROR] Stream '{name}' uses a restricted source (echo/expr/exec) which is disabled by default for security. "
@@ -122,7 +126,7 @@ for name in list(go2rtc_config.get("streams", {})):
                del go2rtc_config["streams"][name]
                continue
            go2rtc_config["streams"][name] = formatted_stream
-        except KeyError as e:
+        except ValueError as e:
            print(
                "[ERROR] Invalid substitution found, see https://docs.frigate.video/configuration/restream#advanced-restream-configurations for more info."
            )
@@ -132,7 +136,7 @@ for name in list(go2rtc_config.get("streams", {})):
        filtered_streams = []
        for i, stream_item in enumerate(stream):
            try:
-                formatted_stream = stream_item.format(**FRIGATE_ENV_VARS)
+                formatted_stream = substitute_frigate_vars(stream_item)
                if is_restricted_go2rtc_source(formatted_stream):
                    print(
                        f"[ERROR] Stream '{name}' item {i + 1} uses a restricted source (echo/expr/exec) which is disabled by default for security. "
@@ -141,7 +145,7 @@ for name in list(go2rtc_config.get("streams", {})):
                    continue

                filtered_streams.append(formatted_stream)
-            except KeyError as e:
+            except ValueError as e:
                print(
                    "[ERROR] Invalid substitution found, see https://docs.frigate.video/configuration/restream#advanced-restream-configurations for more info."
                )
@@ -185,3 +189,6 @@ if config.get("birdseye", {}).get("restream", False):
 # Write go2rtc_config to /dev/shm/go2rtc.yaml
 with open("/dev/shm/go2rtc.yaml", "w") as f:
    yaml.dump(go2rtc_config, f)
+
+# config contains camera credentials; do not leave it world-readable
+os.chmod("/dev/shm/go2rtc.yaml", 0o640)
@@ -0,0 +1,107 @@
+"""Normalize the go2rtc HomeKit file and hand it to go2rtc, as root.
+
+Runs before the drop. The file is in the runtime-user-owned /config, so a
+planted symlink could redirect the root write or chown onto another file;
+every operation goes through an O_NOFOLLOW fd to prevent that.
+
+Usage: prepare_homekit.py PATH [--chown]
+"""
+
+import errno
+import grp
+import io
+import os
+import pwd
+import stat
+import sys
+
+from ruamel.yaml import YAML
+
+RUNTIME_OWNER = "go2rtc"
+SHARED_GROUP = "frigate-data"
+MODE = 0o664
+MAX_BYTES = 10 * 1024 * 1024
+
+
+def open_nofollow(path: str) -> int:
+    """Return an fd to a regular file at path, never following a symlink."""
+    flags = os.O_RDWR | os.O_CREAT | os.O_NOFOLLOW
+    try:
+        fd = os.open(path, flags, MODE)
+    except OSError as err:
+        if err.errno != errno.ELOOP:
+            raise
+        os.unlink(path)
+        return os.open(path, flags | os.O_EXCL, MODE)
+
+    # A fifo or other non-regular file would hang or misbehave on read; replace it.
+    if not stat.S_ISREG(os.fstat(fd).st_mode):
+        os.close(fd)
+        os.unlink(path)
+        return os.open(path, flags | os.O_EXCL, MODE)
+    return fd
+
+
+def normalize(content: str) -> str:
+    """Keep only the homekit section, matching the previous yq/jq behavior."""
+    yaml = YAML(typ="safe")
+    try:
+        data = yaml.load(content)
+    except Exception:
+        return ""
+
+    if not isinstance(data, dict) or "homekit" not in data:
+        return ""
+
+    buf = io.StringIO()
+    yaml.dump({"homekit": data["homekit"]}, buf)
+    return buf.getvalue()
+
+
+def main() -> int:
+    if len(sys.argv) < 2:
+        print("[ERROR] prepare_homekit: PATH is required", file=sys.stderr)
+        return 2
+
+    path = sys.argv[1]
+    do_chown = "--chown" in sys.argv[2:]
+
+    try:
+        fd = open_nofollow(path)
+    except PermissionError:
+        print(
+            f"[WARN] {path} is not writable by uid {os.geteuid()}, so HomeKit "
+            "pairing changes will not persist. It is owned by the go2rtc user "
+            "from an earlier run in the default mode. To fix, on the host run: "
+            f"chown {os.geteuid()}:{os.getegid()} <your config dir>/{os.path.basename(path)}"
+        )
+        return 0
+
+    try:
+        content = os.read(fd, MAX_BYTES).decode("utf-8", "replace")
+        normalized = normalize(content)
+        os.ftruncate(fd, 0)
+        os.lseek(fd, 0, os.SEEK_SET)
+        os.write(fd, normalized.encode("utf-8"))
+
+        if do_chown:
+            # tolerate a chown-refusing mount (NFS root_squash): pairing
+            # persistence degrades, the service does not
+            try:
+                uid = pwd.getpwnam(RUNTIME_OWNER).pw_uid
+                gid = grp.getgrnam(SHARED_GROUP).gr_gid
+                os.fchown(fd, uid, gid)
+                os.fchmod(fd, MODE)
+            except (KeyError, OSError):
+                print(
+                    f"[WARN] Could not hand {path} to the go2rtc user; "
+                    "HomeKit pairing changes may not persist"
+                )
+    finally:
+        os.close(fd)
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -1,9 +1,13 @@
+# Loaded with -c from the /tmp/nginx/conf copy: relative includes follow the -c
+# file, all other path directives follow --prefix and must stay absolute.
+
 daemon off;
+# ignored by a non-root master; keeps workers root under FRIGATE_RUN_AS_ROOT
 user root;
 worker_processes auto;

 error_log /dev/stdout warn;
-pid /var/run/nginx.pid;
+pid /tmp/nginx/nginx.pid;

 events {
    worker_connections 1024;
@@ -11,6 +15,13 @@ events {

 http {
    map_hash_bucket_size 256;
+    server_tokens off;
+
+    client_body_temp_path /tmp/nginx/client_body;
+    proxy_temp_path /tmp/nginx/proxy;
+    fastcgi_temp_path /tmp/nginx/fastcgi;
+    uwsgi_temp_path /tmp/nginx/uwsgi;
+    scgi_temp_path /tmp/nginx/scgi;

    include mime.types;
    default_type application/octet-stream;
@@ -62,6 +73,7 @@ http {

    server {
        include listen.conf;
+        include security_headers.conf;

        # enable HTTP/2 for TLS connections to eliminate browser 6-connection limit
        http2 on;
@@ -75,6 +87,12 @@ http {
        vod_align_segments_to_key_frames on;
        vod_manifest_segment_durations_mode accurate;
        vod_ignore_edit_list on;
+        # short leading segments at each playlist start; sources start at
+        # the seek target, so the ladder applies to every seek. Only
+        # effective when clips declare real keyFrameDurations
+        vod_bootstrap_segment_durations 1000;
+        vod_bootstrap_segment_durations 2000;
+        vod_bootstrap_segment_durations 4000;
        vod_segment_duration 10000;

        # MPEG-TS settings (not used when fMP4 is enabled, kept for reference)
@@ -114,25 +132,23 @@ http {
            # Smaller segments, faster generation, better browser compatibility
            vod_hls_container_format fmp4;

+            # fMP4 playlists use EXT-X-MAP, which requires HLS protocol
+            # version 6 (RFC 8216 section 7); the module default is 4
+            vod_hls_version 6;
+
            secure_token $args;
            secure_token_types application/vnd.apple.mpegurl;

+            include security_headers.conf;
            add_header Cache-Control "no-store";
            expires off;

            keepalive_disable safari;
-
-            # vod module returns 502 for non-existent media
-            # https://github.com/kaltura/nginx-vod-module/issues/468
-            error_page 502 =404 /vod-not-found;
-        }
-
-        location = /vod-not-found {
-            return 404;
        }

        location /stream/ {
            include auth_request.conf;
+            include security_headers.conf;
            add_header Cache-Control "no-store";
            expires off;

@@ -150,10 +166,13 @@ http {
            include auth_request.conf;
            types {
                video/mp4 mp4;
-                image/jpeg jpg;
+                image/jpeg jpg jpeg;
+                image/png png;
+                image/webp webp;
            }

            expires 7d;
+            include security_headers.conf;
            add_header Cache-Control "public";
            autoindex on;
            root /media/frigate;
@@ -246,6 +265,7 @@ http {

        location /api/ {
            include auth_request.conf;
+            include security_headers.conf;
            add_header Cache-Control "no-store";
            expires off;
            proxy_pass http://frigate_api/;
@@ -312,29 +332,27 @@ http {

        location / {
            # do not require auth for static assets
+            include security_headers.conf;
            add_header Cache-Control "no-store";
            expires off;

            location /assets/ {
                access_log off;
                expires 1y;
-                add_header Cache-Control "public";
-            }
-
-            location /fonts/ {
-                access_log off;
-                expires 1y;
+                include security_headers.conf;
                add_header Cache-Control "public";
            }

            location /locales/ {
                access_log off;
+                include security_headers.conf;
                add_header Cache-Control "public";
            }

            location ~ ^/.*-([A-Za-z0-9]+)\.webmanifest$ {
                access_log off;
                expires 1y;
+                include security_headers.conf;
                add_header Cache-Control "public";
                default_type application/json;
                proxy_set_header Accept-Encoding "";
@@ -351,7 +369,7 @@ http {
            sub_filter '"/BASE_PATH/assets/' '"$http_x_ingress_path/assets/';
            sub_filter '"/BASE_PATH/locales/' '"$http_x_ingress_path/locales/';
            sub_filter '"/BASE_PATH/monacoeditorwork/' '"$http_x_ingress_path/assets/';
-            sub_filter 'return"/BASE_PATH/"' 'return window.baseUrl';
+            sub_filter 'return`/BASE_PATH/`' 'return window.baseUrl';
            sub_filter '<body>' '<body><script>window.baseUrl="$http_x_ingress_path/";</script>';
            sub_filter_types text/css application/javascript;
            sub_filter_once off;
@@ -0,0 +1,5 @@
+# Deliberately no X-Frame-Options or CSP frame-ancestors: HA's Webpage card and
+# iframe panels frame Frigate cross-origin, and either would break them
+# silently. Bind-mount this file to add your own.
+add_header X-Content-Type-Options "nosniff" always;
+add_header Referrer-Policy "strict-origin-when-cross-origin" always;
@@ -8,8 +8,8 @@ listen {{ .listen.internal }};
    listen {{ .listen.external }} ssl;
    {{ if .ipv6.enabled }}listen [::]:{{ .listen.external_port }} ssl;{{ end }}

-    ssl_certificate /etc/letsencrypt/live/frigate/fullchain.pem;
-    ssl_certificate_key /etc/letsencrypt/live/frigate/privkey.pem;
+    ssl_certificate {{ .tls.cert_path }}/fullchain.pem;
+    ssl_certificate_key {{ .tls.cert_path }}/privkey.pem;

    # generated 2024-06-01, Mozilla Guideline v5.7, nginx 1.25.3, OpenSSL 1.1.1w, modern configuration, no OCSP
    # https://ssl-config.mozilla.org/#server=nginx&version=1.25.3&config=modern&openssl=1.1.1w&ocsp=false&guideline=5.7
@@ -0,0 +1,55 @@
+#!/bin/bash
+# Ahead-of-time volume ownership migration for switching Frigate to non-root.
+# Run from the host BEFORE enabling PUID/PGID or --user:
+#
+#   ./fix-permissions.sh [--dry-run] <config_dir> <media_dir> [PUID] [PGID]
+#
+# Wraps the image's fix-ownership helper so there is exactly one
+# implementation of the chown logic. Requires an image that contains the
+# helper (any release that includes non-root support).
+
+set -o errexit -o nounset -o pipefail
+
+IMAGE="${FRIGATE_IMAGE:-ghcr.io/blakeblackshear/frigate:stable}"
+
+dry_run_flag=""
+if [[ "${1:-}" == "--dry-run" ]]; then
+    dry_run_flag="--dry-run"
+    shift
+fi
+
+if [[ $# -lt 2 ]]; then
+    echo "Usage: $0 [--dry-run] <config_dir> <media_dir> [PUID] [PGID]" >&2
+    exit 2
+fi
+
+config_dir="$1"
+media_dir="$2"
+puid="${3:-1000}"
+pgid="${4:-1000}"
+
+# The ids are interpolated into the container's bash -c source below, so
+# anything but digits would be reparsed as shell rather than passed through
+if ! [[ "$puid" =~ ^[0-9]+$ && "$pgid" =~ ^[0-9]+$ ]]; then
+    echo "[ERROR] PUID and PGID must be numeric, got '${puid}' and '${pgid}'" >&2
+    exit 2
+fi
+
+echo "[INFO] Using image ${IMAGE} (override with FRIGATE_IMAGE=...)"
+if ! docker image inspect "${IMAGE}" >/dev/null 2>&1; then
+    echo "[INFO] ${IMAGE} is not present locally and has to be pulled first; this may take a while"
+fi
+
+if [[ -n "$dry_run_flag" ]]; then
+    echo "[INFO] Dry run: reporting what would change under ${config_dir} and ${media_dir}, changing nothing"
+else
+    echo "[INFO] Aligning ${config_dir} and ${media_dir} to ${puid}:${pgid}; this may take a while on large filesystems"
+fi
+
+# shellcheck disable=SC2086
+docker run --rm \
+    -v "${config_dir}:/config" \
+    -v "${media_dir}:/media/frigate" \
+    --entrypoint bash \
+    "${IMAGE}" \
+    -c "command -v fix-ownership >/dev/null || { echo '[ERROR] this Frigate image predates non-root support; set FRIGATE_IMAGE to a release that includes it' >&2; exit 1; }; exec fix-ownership ${dry_run_flag} ${puid} ${pgid} /config /media/frigate"
@@ -13,6 +13,16 @@ TRT_VER=${TRT_VER:-$(cat /etc/TENSORRT_VER)}
 OUTPUT_FOLDER="${MODEL_CACHE_DIR}/${TRT_VER}"
 YOLO_MODELS=${YOLO_MODELS:-""}

+# This runs as root after prepare's sentinel-guarded sweep, so the dirs and
+# engines it creates below are the runtime user's to fix up, on every exit path
+function hand_off_ownership() {
+    if [[ "$(id -u)" -eq 0 && "${FRIGATE_RUN_AS_ROOT:-false}" != "true" ]]; then
+        /usr/local/bin/fix-ownership "${PUID:-1000}" "${PGID:-1000}" \
+            /config/model_cache "${MODEL_CACHE_DIR}"
+    fi
+}
+trap hand_off_ownership EXIT
+
 # Create output folder
 mkdir -p ${OUTPUT_FOLDER}

@@ -1,6 +1,5 @@
 # Nvidia ONNX Runtime GPU Support
 --extra-index-url 'https://pypi.nvidia.com'
-cython==3.0.*; platform_machine == 'x86_64'
 nvidia-cuda-cupti-cu12==12.8.90; platform_machine == 'x86_64'
 nvidia-cublas-cu12==12.8.4.1; platform_machine == 'x86_64'
 nvidia-cudnn-cu12==9.8.0.87; platform_machine == 'x86_64'
@@ -56,17 +56,6 @@ mqtt:
  # 2 = exactly once
  qos: 0

-# Optional: Detectors configuration. Defaults to a single CPU detector
-detectors:
-  # Required: name of the detector
-  detector_name:
-    # Required: type of the detector
-    # Frigate provides many types, see https://docs.frigate.video/configuration/object_detectors for more details (default: shown below)
-    # Additional detector types can also be plugged in.
-    # Detectors may require additional configuration.
-    # Refer to the Detectors configuration page for more information.
-    type: cpu
-
 # Optional: Database configuration
 database:
  # The path to store the SQLite DB (default: shown below)
@@ -157,44 +146,56 @@ auth:
      - front_door
      - back_yard

-# Optional: model modifications
+# Optional: object detection models. Defaults to a single model on a CPU detector.
 # NOTE: The default values are for the EdgeTPU detector.
 # Other detectors will require the model config to be set.
-model:
-  # Required: path to the model. Frigate+ models use plus://<model_id> (default: automatic based on detector)
-  path: /edgetpu_model.tflite
-  # Required: path to the labelmap (default: shown below)
-  labelmap_path: /labelmap.txt
-  # Required: Object detection model input width (default: shown below)
-  width: 320
-  # Required: Object detection model input height (default: shown below)
-  height: 320
-  # Required: Object detection model input colorspace
-  # Valid values are rgb, bgr, or yuv. (default: shown below)
-  input_pixel_format: rgb
-  # Required: Object detection model input tensor format
-  # Valid values are nhwc, nchw, hwnc, or hwcn (default: shown below)
-  input_tensor: nhwc
-  # Optional: Data type of the model input tensor
-  # Valid values are float, float_denorm, or int (default: shown below)
-  input_dtype: int
-  # Required: Object detection model architecture, used by detectors that support more
-  # than one model type (openvino, onnx, rknn, memryx, axengine, synaptics, and others)
-  # Valid values are ssd, yolox, yolonas, yolo-generic, rfdetr, dfine (default: shown below)
-  model_type: ssd
-  # Required: Label name modifications. These are merged into the standard labelmap.
-  labelmap:
-    2: vehicle
-  # Optional: Map of object labels to their attribute labels (default: depends on model)
-  attributes_map:
-    person:
-      - amazon
-      - face
-    car:
-      - amazon
-      - fedex
-      - license_plate
-      - ups
+models:
+    # Optional: the camera environment this model is for (default: shown below)
+    # Cameras select a model by setting detect -> scene to a matching value, and
+    # a model with a scene of all is used by any camera that does not set one.
+    # Valid values are all, indoor, outdoor, indoor_thermal, outdoor_thermal
+  - scene: all
+    # Required: hardware this model runs on, as <detector> or <detector>:<device>
+    # See https://docs.frigate.video/configuration/object_detectors for the
+    # detectors available and the devices each one accepts. All of a model's
+    # devices must use the same detector. Listing the same device more than once
+    # runs additional inference processes on it.
+    devices:
+      - edgetpu:pci:0
+    # Required: path to the model. Frigate+ models use plus://<model_id> (default: automatic based on detector)
+    path: /edgetpu_model.tflite
+    # Required: path to the labelmap (default: shown below)
+    labelmap_path: /labelmap.txt
+    # Required: Object detection model input width (default: shown below)
+    width: 320
+    # Required: Object detection model input height (default: shown below)
+    height: 320
+    # Required: Object detection model input colorspace
+    # Valid values are rgb, bgr, or yuv. (default: shown below)
+    input_pixel_format: rgb
+    # Required: Object detection model input tensor format
+    # Valid values are nhwc, nchw, hwnc, or hwcn (default: shown below)
+    input_tensor: nhwc
+    # Optional: Data type of the model input tensor
+    # Valid values are float, float_denorm, or int (default: shown below)
+    input_dtype: int
+    # Required: Object detection model architecture, used by detectors that support more
+    # than one model type (openvino, onnx, rknn, memryx, axengine, synaptics, and others)
+    # Valid values are ssd, yolox, yolonas, yolo-generic, rfdetr, dfine (default: shown below)
+    model_type: ssd
+    # Required: Label name modifications. These are merged into the standard labelmap.
+    labelmap:
+      2: vehicle
+    # Optional: Map of object labels to their attribute labels (default: depends on model)
+    attributes_map:
+      person:
+        - amazon
+        - face
+      car:
+        - amazon
+        - fedex
+        - license_plate
+        - ups

 # Optional: Audio Events Configuration
 # NOTE: Can be overridden at the camera level
@@ -217,6 +218,8 @@ audio:
    - fire_alarm
    - speech
    - yell
+  # Optional: Audio label name modifications. These are merged into the standard audio labelmap.
+  labelmap: {}
  # Optional: Filters to configure detection.
  filters:
    # Label that matches label in listen config.
@@ -251,11 +254,15 @@ birdseye:
  # Optional: Encoding quality of the mpeg1 feed (default: shown below)
  # 1 is the highest quality, and 31 is the lowest. Lower quality feeds utilize less CPU resources.
  quality: 8
-  # Optional: Mode of the view. Available options are: objects, motion, and continuous
-  #   objects - cameras are included if they have had a tracked object within the last 30 seconds
-  #   motion - cameras are included if motion was detected in the last 30 seconds
-  #   continuous - all cameras are included always
-  mode: objects
+  # Optional: Activity types that include cameras in Birdseye (default: shown below)
+  # Multiple activity types can be listed at the same time.
+  #   continuous:  all cameras are included always
+  #   motion:      included if motion was detected within the inactivity threshold
+  #   all_objects: included if a tracked object was present within the inactivity threshold
+  #   alerts:      included while an alert review item is in progress
+  #   detections:  included while a detection review item is in progress
+  modes:
+    - all_objects
  # Optional: Threshold for camera activity to stop showing camera (default: shown below)
  inactivity_threshold: 30
  # Optional: Configure the birdseye layout
@@ -287,6 +294,8 @@ ffmpeg:
    detect: -threads 2 -f rawvideo -pix_fmt yuv420p
    # Optional: output args for record streams (default: shown below)
    record: preset-record-generic
+    # Optional: output args for sub stream record streams (default: the record output args above)
+    # record_sub: preset-record-generic
  # Optional: Time in seconds to wait before ffmpeg retries connecting to the camera. (default: shown below)
  # If set too low, frigate will retry a connection to the camera's stream too frequently, using up the limited streams some cameras can allow at once
  # If set too high, then if a ffmpeg crash or camera stream timeout occurs, you could potentially lose up to a maximum of retry_interval second(s) of footage
@@ -306,6 +315,10 @@ detect:
  width: 1280
  # Optional: height of the frame for the input with the detect role (default: use native stream resolution)
  height: 720
+  # Optional: the environment this camera looks at, which picks the model it runs on
+  # (default: the model with a scene of all)
+  # Valid values are all, indoor, outdoor, indoor_thermal, outdoor_thermal
+  scene: outdoor
  # Optional: desired fps for your camera for the input with the detect role (default: shown below)
  # NOTE: Recommended value of 5. Ideally, try and reduce your FPS on the camera.
  fps: 5
@@ -483,6 +496,11 @@ review:
      - Animals in the garden
    # Optional: Preferred response language (default: English)
    preferred_language: English
+    # Optional: Writing style preset for generated descriptions (default: shown below)
+    # Options: "default", "natural", "concise", "detailed"
+    # Presets adjust the tone and level of detail of the user-facing title,
+    # summary, and scene description; "default" leaves the built-in prompt unchanged.
+    response_style: default
    # Optional: Save thumbnails sent to the GenAI provider for review/debugging purposes (default: shown below)
    debug_save_thumbnails: False

@@ -637,6 +655,42 @@ record:
      #       For example, if the camera retain mode is "motion", the segments without motion are
      #       never stored, so setting the mode to "all" here won't bring them back.
      mode: motion
+  # Optional: Sub stream recording settings
+  # Records a second, lower quality stream for quality selection during playback
+  # and extended low quality retention. Requires the record_sub role to be assigned
+  # to one of the camera's inputs.
+  sub:
+    # Optional: Enable sub stream recording (default: shown below)
+    # NOTE: Recording must also be enabled for sub stream recording to run.
+    enabled: False
+    # Optional: Continuous retention settings for sub stream recordings
+    continuous:
+      # Optional: Number of days to retain sub stream recordings regardless of tracked objects or motion (default: shown below)
+      days: 0
+    # Optional: Motion retention settings for sub stream recordings
+    motion:
+      # Optional: Number of days to retain sub stream recordings triggered by motion (default: shown below)
+      days: 0
+    # Optional: Retention settings for sub stream recordings of alerts
+    # NOTE: Pre and post capture windows are taken from the main alerts config above.
+    alerts:
+      # Required: Retention days (default: shown below)
+      days: 10
+      # Optional: Mode for retention. (default: shown below)
+      #   all - save all sub stream recording segments for alerts regardless of activity
+      #   motion - save all sub stream recording segments for alerts with any detected motion
+      #   active_objects - save all sub stream recording segments for alerts with active/moving objects
+      mode: motion
+    # Optional: Retention settings for sub stream recordings of detections
+    # NOTE: Pre and post capture windows are taken from the main detections config above.
+    detections:
+      # Required: Retention days (default: shown below)
+      days: 10
+      # Optional: Mode for retention. (default: shown below)
+      #   all - save all sub stream recording segments for detections regardless of activity
+      #   motion - save all sub stream recording segments for detections with any detected motion
+      #   active_objects - save all sub stream recording segments for detections with active/moving objects
+      mode: motion

 # Optional: Configuration for the snapshots written to the clips directory for each tracked object
 # Timestamp, bounding_box, crop and height settings are applied by default to API requests for snapshots.
@@ -746,7 +800,7 @@ lpr:
 # to Google or OpenAI's LLMs to generate descriptions. GenAI features can be configured at
 # the camera level to enhance privacy for indoor cameras.
 # NOTE: genai is a map of named providers. Each key is a name you choose for the provider,
-#       and each role (chat, descriptions, embeddings) may be assigned to exactly one provider.
+#       and each role (chat, descriptions, embeddings, transcribe) may be assigned to exactly one provider.
 genai:
  # Required: name of the provider (chosen by you, used to reference it elsewhere)
  my_provider:
@@ -759,11 +813,13 @@ genai:
    # Required: The model to use with the provider.
    model: gemini-1.5-flash
    # Optional: Roles this provider handles (default: shown below)
-    # Each role (chat, descriptions, embeddings) must be assigned to exactly one provider.
+    # Each role (chat, descriptions, embeddings, transcribe) must be assigned to exactly
+    # one provider.
    roles:
      - chat
      - descriptions
      - embeddings
+      - transcribe
    # Optional additional args to pass to the GenAI Provider (default: None)
    provider_options:
      keep_alive: -1
@@ -776,13 +832,19 @@ genai:
 audio_transcription:
  # Optional: Enable live and speech event audio transcription (default: shown below)
  enabled: False
+  # Optional: The transcription backend (default: shown below)
+  # Either 'whisper' for Frigate's built-in local models, or the name of a genai
+  # provider that has 'transcribe' in its roles. device and model_size are ignored
+  # when a genai provider is named.
+  model: whisper
  # Optional: The device to run the models on for live transcription. (default: shown below)
  device: CPU
  # Optional: Set the model size used for live transcription. (default: shown below)
  model_size: small
  # Optional: Set the language used for transcription translation. (default: shown below)
-  # List of language codes: https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10
-  language: en
+  # Use 'auto' to let the model detect the language, or a language code from
+  # https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10
+  language: auto

 # Optional: Configuration for classification models
 classification:
@@ -888,7 +950,7 @@ cameras:
        # Required: the path to the stream
        # NOTE: path may include environment variables or docker secrets, which must begin with 'FRIGATE_' and be referenced in {}
        - path: rtsp://viewer:{FRIGATE_RTSP_PASSWORD}@10.0.10.10:554/cam/realmonitor?channel=1&subtype=2
-          # Required: list of roles for this stream. valid values are: audio,detect,record
+          # Required: list of roles for this stream. valid values are: audio,detect,record,record_sub
          # NOTICE: In addition to assigning the audio, detect, and record roles
          # they must also be enabled in the camera config.
          roles:
@@ -981,7 +1043,9 @@ cameras:
      # Optional: Adjust sort order of cameras in the UI. Larger numbers come later (default: shown below)
      # By default the cameras are sorted alphabetically.
      order: 0
-      # Optional: Whether or not to show the camera in the Frigate UI (default: shown below)
+      # Optional: Whether or not to show the camera on the default All Cameras live dashboard.
+      # The camera is still available everywhere else, including camera groups and settings
+      # (default: shown below)
      dashboard: True
      # Optional: Whether this camera is visible in review (the review page and its camera
      # filter, motion review, and the history view) (default: shown below)
@@ -63,15 +63,9 @@ go2rtc:

 ### `environment_vars`

-This section can be used to set environment variables for those unable to modify the environment of the container, like within Home Assistant OS. Docker users should set environment variables in their `docker run` command (`-e FRIGATE_MQTT_PASSWORD=secret`) or `docker-compose.yml` file (`environment:` section) instead. Note that values set here are stored in plain text in your config file, so if the goal is to keep credentials out of your configuration, use Docker environment variables or Docker secrets instead.
+This section sets environment variables in the Frigate process for those unable to modify the environment of the container, like within Home Assistant OS. It's meant for process settings such as `LIBVA_DRIVER_NAME` or the TensorFlow thread counts below. Docker users should set environment variables in their `docker run` command (`-e LIBVA_DRIVER_NAME=i965`) or `docker-compose.yml` file (`environment:` section) instead. Values set here are stored in plain text in your config file, so credentials belong in `secrets.yaml`, Docker environment variables, or Docker secrets instead.

-Variables prefixed with `FRIGATE_` can be referenced in config fields that support environment variable substitution (such as MQTT host and credentials, camera stream URLs, and ONVIF host and credentials) using the `{FRIGATE_VARIABLE_NAME}` syntax.
-
-:::note
-
-The `go2rtc` section is an exception. go2rtc runs as a separate process, so its stream definitions can only be substituted with variables that exist in the container's environment (set via Docker `-e`, the `environment:` section of `docker-compose.yml`, or Docker secrets). Variables defined in the `environment_vars` block above are not available to go2rtc streams. Home Assistant app users, who cannot set container environment variables, must instead put credentials directly in their go2rtc stream URLs.
-
-:::
+Names prefixed with `FRIGATE_` set here also take part in `{FRIGATE_VARIABLE_NAME}` substitution (see [below](#substitution-sources-and-precedence)), but `secrets.yaml` is the better home for them.

 <ConfigTabs>
 <TabItem value="ui">
@@ -80,23 +74,17 @@ Navigate to <NavPath path="Settings > System > Environment variables" /> to add

 | Field             | Description                                               |
 | ----------------- | --------------------------------------------------------- |
-| **Variable name** | The environment variable name (e.g., `FRIGATE_MQTT_USER`) |
+| **Variable name** | The environment variable name (e.g., `LIBVA_DRIVER_NAME`) |
 | **Value**         | The value for the variable                                |

-Variables defined here can be referenced elsewhere in your configuration using the `{FRIGATE_VARIABLE_NAME}` syntax.
+Names prefixed with `FRIGATE_` can also be referenced elsewhere in your configuration using the `{FRIGATE_VARIABLE_NAME}` syntax.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
 environment_vars:
-  FRIGATE_MQTT_USER: my_mqtt_user
-  FRIGATE_MQTT_PASSWORD: my_mqtt_password
-
-mqtt:
-  host: "{FRIGATE_MQTT_HOST}"
-  user: "{FRIGATE_MQTT_USER}"
-  password: "{FRIGATE_MQTT_PASSWORD}"
+  LIBVA_DRIVER_NAME: i965
 ```

 </TabItem>
@@ -130,6 +118,51 @@ environment_vars:
 </TabItem>
 </ConfigTabs>

+### `secrets.yaml`
+
+A `secrets.yaml` file next to your `config.yml` is an additional source of `FRIGATE_` variables, for installs that can't set container environment variables or mount Docker secrets. It's a flat map of names to values, and it is never read or written by the Frigate UI:
+
+```yaml
+FRIGATE_CAM_USER: viewer
+FRIGATE_CAM_PASS: "p@ss w0rd"
+FRIGATE_MQTT_HOST: mqtt.internal.example
+```
+
+For Docker this is `/config/secrets.yaml` inside the container, so it lives in whatever host directory you mounted at `/config`. For the Home Assistant App it's `/addon_configs/<addon_directory>/secrets.yaml`, in the same folder as your `config.yml`; see [the App config directory](../config.md#accessing-app-config-dir) for the directory name for your variant.
+
+Names must start with `FRIGATE_`, and nesting is not supported. `secrets.yaml` feeds `{FRIGATE_VARIABLE_NAME}` substitution, so the handful of variables Frigate reads straight from the process environment, such as `FRIGATE_JWT_SECRET`, still need a container environment variable or a Docker secret.
+
+### Substitution sources and precedence
+
+The same `{FRIGATE_VARIABLE_NAME}` placeholder resolves from four sources. When a name is defined in more than one, the higher one wins and a warning at startup names which source was used.
+
+| Priority    | Source                | Where it's set                                                             | Who can use it                 |
+| ----------- | --------------------- | -------------------------------------------------------------------------- | ------------------------------ |
+| 1 (highest) | Docker secrets        | Files in `/run/secrets`, or the directory named by `CREDENTIALS_DIRECTORY` | Docker, systemd                |
+| 2           | Container environment | `docker run -e`, the `environment:` section of `docker-compose.yml`        | Docker                         |
+| 3           | `secrets.yaml`        | Next to `config.yml`, see above                                            | Everyone, including the HA App |
+| 4 (lowest)  | `environment_vars`    | The block in `config.yml` described above                                  | Everyone, including the HA App |
+
+For example, with this `secrets.yaml`:
+
+```yaml
+FRIGATE_MQTT_PASSWORD: from_secrets
+```
+
+and this `config.yml`:
+
+```yaml
+environment_vars:
+  FRIGATE_MQTT_PASSWORD: from_config
+
+mqtt:
+  password: "{FRIGATE_MQTT_PASSWORD}"
+```
+
+the password resolves to `from_secrets`, and the log shows `FRIGATE_MQTT_PASSWORD is defined in more than one place, using the value from secrets.yaml`. Add `-e FRIGATE_MQTT_PASSWORD=from_env` to the container and it resolves to `from_env` instead.
+
+Referencing a name that no source defines is a config validation error naming the field.
+
 ### `database`

 Tracked object and recording information is managed in a sqlite database at `/config/frigate.db`. If that database is deleted, recordings will be orphaned and will need to be cleaned up manually. They also won't show up in the Media Browser within Home Assistant.
@@ -177,7 +210,7 @@ Custom models may also require different input tensor formats. The colorspace co
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and open the **Custom Model** tab to configure the model path, dimensions, and input format.
+Navigate to <NavPath path="Settings > System > Detection models" /> and, on the model you want to change, open the **Custom Model** tab to configure the model path, dimensions, and input format.

 | Field                                         | Description                          |
 | --------------------------------------------- | ------------------------------------ |
@@ -192,12 +225,14 @@ Navigate to <NavPath path="Settings > System > Detectors and model" /> and open

 ```yaml
 # Optional: model config
-model:
-  path: /path/to/model
-  width: 320
-  height: 320
-  input_tensor: "nhwc"
-  input_pixel_format: "bgr"
+models:
+  - devices:
+      - openvino:GPU
+    path: /path/to/model
+    width: 320
+    height: 320
+    input_tensor: "nhwc"
+    input_pixel_format: "bgr"
 ```

 </TabItem>
@@ -214,15 +249,15 @@ If the labelmap is customized then the labels used for alerts will need to be ad
 The labelmap can be customized to your needs. A common reason to do this is to combine multiple object types that are easily confused when you don't need to be as granular such as car/truck. By default, truck is renamed to car because they are often confused. You cannot add new object types, but you can change the names of existing objects in the model.

 ```yaml
-model:
-  labelmap:
-    2: vehicle
-    3: vehicle
-    5: vehicle
-    7: vehicle
-    15: animal
-    16: animal
-    17: animal
+models:
+  - labelmap:
+      2: vehicle
+      3: vehicle
+      5: vehicle
+      7: vehicle
+      15: animal
+      16: animal
+      17: animal
 ```

 Note that if you rename objects in the labelmap, you will also need to update your `objects -> track` list as well.
@@ -293,6 +328,10 @@ networking:

 This setting is for advanced users. For the majority of use cases it's recommended to change the `ports` section of your Docker compose file or use the Docker `run` `--publish` option instead, e.g. `-p 443:8971`. Changing Frigate's ports may break some integrations.

+The internal and external ports must be different port numbers, and Frigate will refuse to start otherwise. Requests arriving on the internal port are treated as authenticated admins, so pointing both at the same port would remove authentication from the external one.
+
+Nginx binds these ports when it starts, so port changes only take effect after Frigate restarts.
+
 :::

 ### Customizing the Nginx configuration
@@ -358,6 +397,10 @@ To do this:
 2. Update the `ffmpeg.path` in your Frigate config to `/config/custom-ffmpeg`.
 3. Restart Frigate and the custom version will be used if the steps above were done correctly.

+Both binaries have to be executable by Frigate's unprivileged runtime user, so `chmod 755` them after extracting. The startup ownership sweep runs only once, so anything you add to `/config` later keeps whatever ownership and mode you gave it.
+
+There is one exception, and it only affects [`FRIGATE_ROOT_SERVICES`](/configuration/non_root#keeping-individual-services-root) listing `frigate`. That mode runs Frigate as root while still handing `/config` to the unprivileged runtime user, so anything running as that user could swap the binary and gain root. A build inside any of Frigate's writable volumes (`/config`, `/media/frigate`, the cache and shm dirs) is ignored there and the bundled one is used, with a warning in the log. Keep the build somewhere root-owned (any absolute `ffmpeg.path` works, so a read-only bind mount such as `/opt/custom-ffmpeg` is enough) if you need both. The default mode and `FRIGATE_RUN_AS_ROOT=true` are unaffected and behave exactly as they always have.
+
 ### Custom go2rtc version

 Frigate currently includes go2rtc v1.9.14, there may be certain cases where you want to run a different version of go2rtc.
@@ -366,9 +409,11 @@ To do this:

 1. Download the go2rtc build to the `/config` folder.
 2. Rename the build to `go2rtc`.
-3. Give `go2rtc` execute permission.
+3. Give `go2rtc` execute permission for all users (`chmod 755`). It runs as its own `go2rtc` user, which doesn't own the file, so owner-only execute permission isn't enough.
 4. Restart Frigate and the custom version will be used, you can verify by checking go2rtc logs.

+The same exception applies, and again only to [`FRIGATE_ROOT_SERVICES`](/configuration/non_root#keeping-individual-services-root) listing `go2rtc`: the binary is ignored there and the embedded one is used, with a warning in the log. Unlike `ffmpeg.path`, the go2rtc binary location is not configurable, so there is no outside-`/config` alternative. Use `FRIGATE_RUN_AS_ROOT=true` instead if you need both a custom go2rtc build and root. The default mode and the escape hatch both honor `/config/go2rtc` exactly as they always have.
+
 ## Validating your config.yml file updates

 When frigate starts up, it checks whether your config file is valid, and if it is not, the process exits. To minimize interruptions when updating your config, you have three options -- you can edit the config via the WebUI which has built in validation, use the config API, or you can validate on the command line using the frigate docker container.
@@ -114,6 +114,30 @@ audio:
 </TabItem>
 </ConfigTabs>

+#### Grouping Audio Labels
+
+Related audio classes can be grouped under one label by mapping their numeric
+class IDs to the same name. Add the grouped name to `listen` and use it for any
+corresponding filter:
+
+```yaml
+audio:
+  listen:
+    - dogs
+  labelmap:
+    69: dogs # dog
+    70: dogs # bark
+    75: dogs # whimper_dog
+  filters:
+    dogs:
+      threshold: 0.8
+```
+
+Class IDs are zero-based indices in
+[`audio-labelmap.txt`](https://github.com/blakeblackshear/frigate/blob/dev/audio-labelmap.txt),
+so each ID is one less than the displayed file line number.
+Audio label mappings are separate from the object detector's `model.labelmap`.
+
 ### Common Audio Labels

 The labelmap includes hundreds of sound types. The labels below are the ones most users may find practical, grouped by what they're typically used for. Use the exact label string from the left column in your `listen` config, or search for the label in the Frigate UI directly.
@@ -180,7 +204,7 @@ Frequently-heard labels like `speech` can generate a lot of events, and each eve

 ### Audio Transcription

-Frigate supports fully local audio transcription using either `sherpa-onnx` or OpenAI's open-source Whisper models via `faster-whisper`. The goal of this feature is to support Semantic Search for `speech` audio events. Frigate is not intended to act as a continuous, fully-automatic speech transcription service. Automatically transcribing all speech (or queuing many audio events for transcription) requires substantial CPU (or GPU) resources and is impractical on most systems. For this reason, transcriptions for events are initiated manually from the UI or the API rather than being run continuously in the background.
+Frigate supports fully local audio transcription using either `sherpa-onnx` or OpenAI's open-source Whisper models via `faster-whisper`, and can alternatively offload transcription to a [GenAI provider](#genai-provider). The goal of this feature is to support Semantic Search for `speech` audio events. Frigate is not intended to act as a continuous, fully-automatic speech transcription service. Automatically transcribing all speech (or queuing many audio events for transcription) requires substantial CPU (or GPU) resources and is impractical on most systems. For this reason, transcriptions for events are initiated manually from the UI or the API rather than being run continuously in the background.

 :::info

@@ -200,6 +224,7 @@ To enable transcription, configure it globally and optionally disable for specif
 **Global:** Navigate to <NavPath path="Settings > Enrichments > Audio transcription" />.

 - Set **Enable audio transcription** to on
+- Set **Audio transcription model or GenAI provider name** to `whisper` for Frigate's built-in local models, or to the name of a GenAI provider
 - Set **Transcription device** to the desired device
 - Set **Model size** to the desired size

@@ -211,6 +236,7 @@ To enable transcription, configure it globally and optionally disable for specif
 ```yaml
 audio_transcription:
  enabled: True
+  model: whisper
  device: ...
  model_size: ...
 ```
@@ -239,24 +265,92 @@ The optional config parameters that can be set at the global level include:
 - **`enabled`**: Enable or disable the audio transcription feature.
  - Default: `False`
  - It is recommended to only configure the features at the global level, and enable it at the individual camera level.
+- **`model`**: The transcription backend.
+  - Default: `whisper`
+  - `whisper` uses Frigate's built-in local models, described by `device` and `model_size` below.
+  - Any other value must name a key in your `genai` config whose entry has `transcribe` in its `roles`. See [GenAI Provider](#genai-provider).
 - **`device`**: Device to use to run transcription and translation models.
  - Default: `CPU`
  - This can be `CPU` or `GPU`. The `sherpa-onnx` models are lightweight and run on the CPU only. The `whisper` models can run on GPU but are only supported on CUDA hardware.
+  - Ignored when `model` names a GenAI provider.
 - **`model_size`**: The size of the model used for live transcription.
  - Default: `small`
  - This can be `small` or `large`. The `small` setting uses `sherpa-onnx` models that are fast, lightweight, and always run on the CPU but are not as accurate as the `whisper` model.
-  - This config option applies to **live transcription only**. Recorded `speech` events will always use a different `whisper` model (and can be accelerated for CUDA hardware if available with `device: GPU`).
- **`language`**: Defines the language used by `whisper` to translate `speech` audio events (and live audio only if using the `large` model).
-  - Default: `en`
-  - You must use a valid [language code](https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10).
+  - This config option applies to **live transcription only**. With `model: whisper`, recorded `speech` events always use a different `whisper` model (and can be accelerated for CUDA hardware if available with `device: GPU`).
+  - Ignored when `model` names a GenAI provider.
+- **`language`**: Defines the language used to transcribe and translate `speech` audio events (and live audio only if using the `large` model or a GenAI provider).
+  - Default: `auto`
+  - `auto` lets the model detect the language itself, which most models do well. Set an explicit language only if detection is picking the wrong one.
+  - Otherwise you must use a valid [language code](https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10).
  - Transcriptions for `speech` events are translated.
  - Live audio is translated only if you are using the `large` model. The `small` `sherpa-onnx` model is English-only.

-The only field that is valid at the camera level is `enabled`.
+The only field that is valid at the camera level is `enabled`. In particular `model` is global only: the transcription backend is a process-wide resource shared by every camera.
+
+#### GenAI Provider
+
+Frigate can send audio to a GenAI provider for transcription when that provider has the `transcribe` role. This is useful if you already run a GenAI provider, or if you do not have the CPU/GPU headroom for a local whisper model. Supported providers are **OpenAI**, **Azure OpenAI**, **Gemini**, and **llama.cpp** with an audio-capable model (a dedicated ASR model such as Qwen3-ASR, or a general multimodal model that accepts audio). Ollama is not supported as it has no audio input.
+
+To use a GenAI provider for audio transcription:
+
+1. Configure a GenAI provider with `transcribe` in its `roles`.
+2. Set the audio transcription model to that GenAI config key (e.g. `whisper_cloud`).
+
+<ConfigTabs>
+<TabItem value="ui">
+
+Navigate to <NavPath path="Settings > Enrichments > Audio transcription" />.
+
+| Field                                                | Description                                                                                              |
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
+| **Audio transcription model or GenAI provider name** | Set to the GenAI config key (e.g. `whisper_cloud`) to use a configured GenAI provider for transcription |
+
+The GenAI provider must also be configured with the `transcribe` role under <NavPath path="Settings > Enrichments > Generative AI" />.
+
+</TabItem>
+<TabItem value="yaml">
+
+```yaml
+genai:
+  whisper_cloud:
+    provider: openai
+    api_key: your-api-key
+    model: gpt-transcribe
+    roles:
+      - transcribe
+
+audio_transcription:
+  enabled: True
+  model: whisper_cloud
+  language: en
+```
+
+</TabItem>
+</ConfigTabs>
+
+:::warning
+
+**Give `transcribe` its own `genai` entry.** A `genai` entry has a single `model` string that is shared by every role it holds, so `roles: [descriptions, transcribe]` would send the same model name to both the chat endpoint and the transcription endpoint. Transcription models and chat models are almost never the same model, so define a dedicated entry as shown above.
+
+:::
+
+:::warning
+
+**Live transcription against a metered provider is billed continuously.** In live mode Frigate uploads an overlapping ~2 second window of audio roughly once per second, per camera, for as long as audio stays above that camera's `audio.min_volume`. Windows below that threshold are never uploaded, which is what keeps a quiet camera near zero requests, but a camera pointed at a busy street will keep sending.
+
+Three things keep this opt-in: `transcribe` is not one of the default roles, live transcription is off by default, and the volume gate suppresses silence. Transcription of recorded `speech` events is unaffected - it remains a manual, one-request-per-event action.
+
+:::
+
+`device` and `model_size` have no effect on this path and no local model is ever downloaded.
+
+`language` defaults to `auto`, which sends no language hint and lets the model detect it. Most audio models detect language well, so leave it on `auto` unless detection is picking the wrong one.
+
+When set explicitly, it is sent as the transcription endpoint's native `language` parameter for OpenAI, Azure, and llama.cpp, and as part of the prompt for Gemini. This matters for dedicated ASR models such as Qwen3-ASR: they read the prompt as contextual biasing rather than as an instruction, so a language named in the prompt is ignored, while the endpoint parameter is honored.

 #### Live transcription

-The single camera Live view in the Frigate UI supports live transcription of audio for streams defined with the `audio` role. Use the Enable/Disable Live Audio Transcription button/switch to toggle transcription processing. When speech is heard, the UI will display a black box over the top of the camera stream with text. The MQTT topic `frigate/<camera_name>/audio/transcription` will also be updated in real-time with transcribed text.
+The single camera Live view in the Frigate UI supports live transcription of audio for streams defined with the `audio` role. Use the Enable/Disable Live Audio Transcription button/switch to toggle transcription processing, or toggle it outside of the UI with the [`frigate/<camera_name>/audio_transcription/set`](/integrations/mqtt#frigatecamera_nameaudio_transcriptionset) MQTT topic or the HTTP API. When speech is heard, the UI will display a black box over the top of the camera stream with text. The MQTT topic `frigate/<camera_name>/audio/transcription` will also be updated in real-time with transcribed text.

 Results can be error-prone due to a number of factors, including:

@@ -268,6 +362,8 @@ Results can be error-prone due to a number of factors, including:

 For speech sources close to the camera with minimal background noise, use the `small` model.

+A [GenAI provider](#genai-provider) is generally the most accurate option for live transcription, at the cost of a network round trip per window. That round trip has to stay under about a second to keep up with the audio; if it does not, Frigate drops the oldest buffered audio rather than letting the backlog grow.
+
 If you have CUDA hardware, you can experiment with the `large` `whisper` model on GPU. Performance is not quite as fast as the `sherpa-onnx` `small` model, but live transcription is far more accurate. Using the `large` model with CPU will likely be too slow for real-time transcription.

 #### Transcription and translation of `speech` audio events
@@ -284,7 +380,7 @@ Only one `speech` event may be transcribed at a time. Frigate does not automatic

 :::

-Recorded `speech` events will always use a `whisper` model, regardless of the `model_size` config setting. Without a supported Nvidia GPU, generating transcriptions for longer `speech` events may take a fair amount of time, so be patient.
+With `model: whisper`, recorded `speech` events always use a `whisper` model, regardless of the `model_size` config setting. Without a supported Nvidia GPU, generating transcriptions for longer `speech` events may take a fair amount of time, so be patient. With a [GenAI provider](#genai-provider), the recorded clip is sent to the provider instead and no local model is used.

 #### FAQ

@@ -22,7 +22,9 @@ The following ports are available to access the Frigate web UI.

 ## Onboarding

-On startup, an admin user and password are generated and printed in the logs. It is recommended to set a new password for the admin account after logging in for the first time under Settings > Users.
+On startup, an admin user and password are generated and printed in the logs. It is recommended to set a new password for the admin account after logging in for the first time.
+
+On a new install the [setup wizard](../guides/getting_started.md#configuring-frigate) offers this as its first step, along with creating accounts for anyone else who needs access. You can also do both at any time under <NavPath path="Settings > Users" />.

 ## Resetting admin password

@@ -214,9 +216,9 @@ A default role can be provided. Any value in the mapped `role` header will overr

 Navigate to <NavPath path="Settings > System > Proxy" /> and set the default role.

-| Field            | Description                                                   |
-| ---------------- | ------------------------------------------------------------- |
-| **Default role** | Fallback role when no role header is present (e.g., `viewer`) |
+| Field            | Description                                                                                          |
+| ---------------- | ---------------------------------------------------------------------------------------------------- |
+| **Default role** | Fallback role when no role header is present (e.g., `viewer`), or `None (deny access)` to reject unmapped users |

 </TabItem>
 <TabItem value="yaml">
@@ -230,6 +232,14 @@ proxy:
 </TabItem>
 </ConfigTabs>

+Setting `default_role` to `none` denies access instead of falling back to a role. Any proxy-authenticated user whose headers do not match an explicit `role_map` entry receives a 403 response. This is useful when the upstream proxy authenticates a broader set of users than should reach Frigate, so that only mapped groups are allowed in.
+
+```yaml
+proxy:
+  ...
+  default_role: none
+```
+
 ## Role mapping

 In some environments, upstream identity providers (OIDC, SAML, LDAP, etc.) do not pass a Frigate-compatible role directly, but instead pass one or more group claims. To handle this, Frigate supports a `role_map` that translates upstream group names into Frigate's internal roles (`admin`, `viewer`, or custom). This is configurable via YAML in the configuration file:
@@ -255,7 +265,7 @@ In this example:
 - If the proxy passes a role header containing `sysadmins` or `access-level-security`, the user is assigned the `admin` role.
 - If the proxy passes a role header containing `camera-viewer`, the user is assigned the `viewer` role.
 - If the proxy passes a role header containing `operators`, the user is assigned the `operator` custom role.
- If no mapping matches, Frigate falls back to `default_role` if configured.
+- If no mapping matches, Frigate falls back to `default_role` if configured, or denies access if `default_role` is `none`.
 - If `role_map` is not defined, Frigate assumes the role header directly contains `admin`, `viewer`, or a custom role name.

 **Note on matching semantics:**
@@ -329,7 +339,7 @@ Frigate supports user roles to control access to certain features in the UI and

 - **admin**: Full access to all features, including user management and configuration.
 - **viewer**: Read-only access to the UI and API, including viewing cameras, review items, and historical footage. Configuration editor and settings in the UI are inaccessible.
- **Custom Roles**: Arbitrary role names (alphanumeric, dots/underscores) with specific camera permissions. These extend the system for granular access (e.g., "operator" for select cameras).
+- **Custom Roles**: Arbitrary role names (alphanumeric, dots/underscores) with specific camera permissions. These extend the system for granular access (e.g., "operator" for select cameras). The names `admin`, `viewer`, and `none` are reserved and cannot be used.

 ### Custom Roles and Camera Access

@@ -18,13 +18,17 @@ Each camera tile in Birdseye is composed from the frames of the stream assigned

 ## Birdseye Behavior

-### Birdseye Modes
+### Birdseye Activity Types

-Birdseye offers different modes to customize which cameras show under which circumstances.
+Birdseye offers independent activity types that control when cameras are shown. Multiple activity types can be listed together.

- **continuous:** All cameras are always included
- **motion:** Cameras that have detected motion within the last 30 seconds are included
- **objects:** Cameras that have tracked an active object within the last 30 seconds are included
+- **continuous:** The camera is always included
+- **motion:** The camera is included when motion was detected within the last 30 seconds
+- **all_objects:** The camera is included when a tracked object is present, active or stationary
+- **alerts:** The camera is included while an alert review item is in progress
+- **detections:** The camera is included while a detection review item is in progress
+
+`alerts` and `detections` follow the review item's own lifetime, so the camera is removed as soon as the review item ends. Which objects qualify for each is set in [review configuration](./review.md).

 ### Custom Birdseye Icon

@@ -39,27 +43,29 @@ To include a camera in Birdseye view only for specific circumstances, or exclude

 **Global settings:** Navigate to <NavPath path="Settings > System > Birdseye" /> to configure the default Birdseye behavior for all cameras.

-**Per-camera overrides:** Navigate to <NavPath path="Settings > Camera configuration > Birdseye" /> to override the mode or disable Birdseye for a specific camera.
+**Per-camera overrides:** Navigate to <NavPath path="Settings > Camera configuration > Birdseye" /> to override the activity types or disable Birdseye for a specific camera.

-| Field               | Description                                                   |
-| ------------------- | ------------------------------------------------------------- |
-| **Enable Birdseye** | Whether this camera appears in Birdseye view                  |
-| **Tracking mode**   | When to show the camera: `continuous`, `motion`, or `objects` |
+| Field                  | Description                                                |
+| ---------------------- | ---------------------------------------------------------- |
+| **Enable Birdseye**    | Whether this camera appears in Birdseye view               |
+| **Activity types**     | Conditions that determine when to show the camera          |

 </TabItem>
 <TabItem value="yaml">

-```yaml {8-10,12-14}
+```yaml {10-12,15-16}
 # Include all cameras by default in Birdseye view
 birdseye:
  enabled: True
-  mode: continuous
+  modes:
+    - continuous

 cameras:
  front:
-    # Only include the "front" camera in Birdseye view when objects are detected
+    # Only include the "front" camera in Birdseye view when an alert is in progress
    birdseye:
-      mode: objects
+      modes:
+        - alerts
  back:
    # Exclude the "back" camera from Birdseye view
    birdseye:
@@ -71,7 +77,7 @@ cameras:

 ### Birdseye Inactivity

-By default birdseye shows all cameras that have had the configured activity in the last 30 seconds. This threshold can be configured.
+By default birdseye shows all cameras that have had the configured activity in the last 30 seconds. This threshold can be configured, and applies to the `motion` and `all_objects` activity types only.

 <ConfigTabs>
 <TabItem value="ui">
@@ -140,7 +146,8 @@ Navigate to <NavPath path="Settings > System > Birdseye" /> and in the **Camera
 # Include all cameras by default in Birdseye view
 birdseye:
  enabled: True
-  mode: continuous
+  modes:
+    - continuous

 cameras:
  front:
@@ -9,7 +9,7 @@ import NavPath from "@site/src/components/NavPath";

 ## Adding a camera with the Add Camera Wizard

-The Add Camera Wizard is the recommended way to add a camera. Click **Add Camera** in <NavPath path="Settings > Global configuration > Camera management" />. The wizard connects to your camera, tests each stream, and writes the camera's configuration for you, including the [go2rtc](go2rtc.md) restream and the live view stream mapping, so a standard setup needs no hand-written YAML.
+The Add Camera Wizard is the recommended way to add a camera. Click **Add Camera** in <NavPath path="Settings > Global configuration > Camera management" />, or use it from the [setup wizard](../guides/getting_started.md#configuring-frigate) on a new install. The wizard connects to your camera, tests each stream, and writes the camera's configuration for you, including the [go2rtc](go2rtc.md) restream and the live view stream mapping, so a standard setup needs no hand-written YAML.

 ### Step 1: Name and connection

@@ -50,6 +50,31 @@ Connect each stream to get a live preview, an estimated bandwidth figure, and a

 Other features, including [hardware acceleration](hardware_acceleration_video.md), [two way talk](/configuration/live#two-way-talk), and audio transcoding, is configured after the camera has been added. For camera model specific quirks, see the [camera specific](camera_specific.md) docs.

+## Deleting a camera
+
+Click **Delete Camera** in <NavPath path="Settings > Global configuration > Camera management" />, choose the camera, and confirm. Deleting a camera requires the `admin` role and cannot be undone.
+
+:::warning
+
+Deleting a camera permanently removes its recordings, tracked objects, and configuration. If you only want to stop processing a camera, set its state to **Off** or **Disabled** in <NavPath path="Settings > Global configuration > Camera management" /> instead. See [camera state](/configuration/live#camera-state).
+
+:::
+
+Deleting a camera removes:
+
+- The camera's section of your config file, along with its entries in any [role](authentication.md#user-roles) camera list. A custom role left with no cameras is removed as well.
+- Every database record for the camera: tracked objects, review items, recordings, previews, timeline entries, the saved region grid, and [triggers](semantic_search.md#triggers).
+- Every media file for the camera: recordings, snapshots, thumbnails, and preview clips.
+
+[Exports](/usage/exports) are kept by default, so saved footage survives the deletion of the camera it came from. Turn on **Also delete exports for this camera** in the confirmation step to remove those too.
+
+The camera's processes are stopped and the change takes effect immediately, so no restart is required. If the resulting config cannot be parsed, Frigate restores the previous config and reports an error instead of leaving Frigate in a broken state.
+
+Two things are not cleaned up for you:
+
+- **go2rtc streams.** Frigate makes a best effort to stop a running [go2rtc](go2rtc.md) stream named after the camera, but stream entries in your config file remain and are recreated on the next restart. Remove them in <NavPath path="Settings > System > go2rtc streams" /> or in your config file.
+- **Camera groups.** A deleted camera stays listed in any [camera group](#setting-up-camera-groups) that referenced it. The group skips the missing camera, so this is harmless, but you can edit the group to drop the stale entry.
+
 ## Setting Up Camera Inputs

 Several inputs can be configured for each camera and the role of each input can be mixed and matched based on your needs. This allows you to use a lower resolution stream for object detection, but create recordings from a higher resolution stream, or vice versa.
@@ -58,11 +83,12 @@ A camera is enabled by default but can be disabled by using `enabled: False`. Ca

 Each role can only be assigned to one input per camera. The options for roles are as follows:

-| Role     | Description                                                                         |
-| -------- | ----------------------------------------------------------------------------------- |
-| `detect` | Main feed for object detection. [docs](object_detectors.md)                         |
-| `record` | Saves segments of the video feed based on configuration settings. [docs](record.md) |
-| `audio`  | Feed for audio based detection. [docs](audio_detectors.md)                          |
+| Role         | Description                                                                                                  |
+| ------------ | ------------------------------------------------------------------------------------------------------------ |
+| `detect`     | Main feed for object detection. [docs](object_detectors.md)                                                  |
+| `record`     | Saves segments of the video feed based on configuration settings. [docs](record.md)                          |
+| `record_sub` | Saves segments of a second, lower quality stream with its own retention. [docs](record.md#sub-stream-recording) |
+| `audio`      | Feed for audio based detection. [docs](audio_detectors.md)                                                   |

 <ConfigTabs>
 <TabItem value="ui">
@@ -100,7 +100,7 @@ VS Code supports JSON schemas for automatically validating configuration files.

 ## Environment Variable Substitution

-Frigate supports the use of environment variables starting with `FRIGATE_` **only** where specifically indicated in the [reference config](./advanced/reference.md). For example, the following values can be replaced at runtime by using environment variables:
+Frigate supports the use of environment variables starting with `FRIGATE_` **only** where specifically indicated in the [reference config](./advanced/reference.md). See [substitution sources and precedence](./advanced/system.md#substitution-sources-and-precedence) for where those values can come from, including `secrets.yaml`. For example, the following values can be replaced at runtime by using environment variables:

 ```yaml
 mqtt:
@@ -154,7 +154,7 @@ Here are some common starter configuration examples. These can be configured thr

 1. Navigate to <NavPath path="Settings > System > MQTT" /> and configure the MQTT connection to your Home Assistant Mosquitto broker
 2. Navigate to <NavPath path="Settings > Global configuration > FFmpeg" /> and set **Hardware acceleration arguments** to `Raspberry Pi (H.264)`
-3. Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a detector with **Type** `EdgeTPU` and **Device** `usb`
+3. Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown
 4. Navigate to <NavPath path="Settings > Global configuration > Recording" /> and set **Enable recording** to on, **Motion retention > Retention days** to `7`, **Alert retention > Event retention > Retention days** to `30`, **Alert retention > Event retention > Retention mode** to `motion`, **Detection retention > Event retention > Retention days** to `30`, **Detection retention > Event retention > Retention mode** to `motion`
 5. Navigate to <NavPath path="Settings > Global configuration > Snapshots" /> and set **Enable snapshots** to on, **Snapshot retention > Default retention** to `30`
 6. Navigate to <NavPath path="Settings > Global configuration > Camera management" /> and add your camera with the appropriate RTSP stream URL
@@ -172,10 +172,9 @@ mqtt:
 ffmpeg:
  hwaccel_args: preset-rpi-64-h264

-detectors:
-  coral:
-    type: edgetpu
-    device: usb
+models:
+  - devices:
+      - edgetpu:usb

 record:
  enabled: True
@@ -233,7 +232,7 @@ cameras:

 1. Navigate to <NavPath path="Settings > System > MQTT" /> and set **Enable MQTT** to off
 2. Navigate to <NavPath path="Settings > Global configuration > FFmpeg" /> and set **Hardware acceleration arguments** to `VAAPI (Intel/AMD GPU)`
-3. Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a detector with **Type** `EdgeTPU` and **Device** `usb`
+3. Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown
 4. Navigate to <NavPath path="Settings > Global configuration > Recording" /> and set **Enable recording** to on, **Motion retention > Retention days** to `7`, **Alert retention > Event retention > Retention days** to `30`, **Alert retention > Event retention > Retention mode** to `motion`, **Detection retention > Event retention > Retention days** to `30`, **Detection retention > Event retention > Retention mode** to `motion`
 5. Navigate to <NavPath path="Settings > Global configuration > Snapshots" /> and set **Enable snapshots** to on, **Snapshot retention > Default retention** to `30`
 6. Navigate to <NavPath path="Settings > Global configuration > Camera management" /> and add your camera with the appropriate RTSP stream URL
@@ -249,10 +248,9 @@ mqtt:
 ffmpeg:
  hwaccel_args: preset-vaapi

-detectors:
-  coral:
-    type: edgetpu
-    device: usb
+models:
+  - devices:
+      - edgetpu:usb

 record:
  enabled: True
@@ -310,8 +308,8 @@ cameras:

 1. Navigate to <NavPath path="Settings > System > MQTT" /> and configure the connection to your MQTT broker
 2. Navigate to <NavPath path="Settings > Global configuration > FFmpeg" /> and set **Hardware acceleration arguments** to `VAAPI (Intel/AMD GPU)`
-3. Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a detector with **Type** `openvino` and **Device** `AUTO`
-4. On the same page, in the **Custom Model** tab, configure the OpenVINO model path and settings
+3. Navigate to <NavPath path="Settings > System > Detection models" /> and select **Intel GPU** from the **Hardware** dropdown
+4. On the same model, open the **Custom Model** tab and configure the OpenVINO model path and settings
 5. Navigate to <NavPath path="Settings > Global configuration > Recording" /> and set **Enable recording** to on, **Motion retention > Retention days** to `7`, **Alert retention > Event retention > Retention days** to `30`, **Alert retention > Event retention > Retention mode** to `motion`, **Detection retention > Event retention > Retention days** to `30`, **Detection retention > Event retention > Retention mode** to `motion`
 6. Navigate to <NavPath path="Settings > Global configuration > Snapshots" /> and set **Enable snapshots** to on, **Snapshot retention > Default retention** to `30`
 7. Navigate to <NavPath path="Settings > Global configuration > Camera management" /> and add your camera with the appropriate RTSP stream URL
@@ -329,15 +327,12 @@ mqtt:
 ffmpeg:
  hwaccel_args: preset-vaapi

-detectors:
-  ov:
-    type: openvino
-    device: AUTO
-
-model:
-  width: 300
-  height: 300
-  input_tensor: nhwc
+models:
+  - devices:
+      - openvino:AUTO
+    width: 300
+    height: 300
+    input_tensor: nhwc
  input_pixel_format: bgr
  path: /openvino-model/ssdlite_mobilenet_v2.xml
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt
@@ -11,7 +11,7 @@ Object classification allows you to train a custom MobileNetV2 classification mo

 :::info

-Training a custom object classification model requires a one-time internet connection to download MobileNetV2 base weights. Once trained, the model runs fully offline. See [Network Requirements](/frigate/network_requirements#one-time-model-downloads) for details.
+Training a custom object classification model requires an internet connection to download MobileNetV2 base weights. By default these weights are not cached in `/config/`, so they are downloaded again after the container is recreated. Once trained, the model runs fully offline. See [Network Requirements](/frigate/network_requirements#one-time-model-downloads) for details.

 :::

@@ -11,7 +11,7 @@ State classification allows you to train a custom MobileNetV2 classification mod

 :::info

-Training a custom state classification model requires a one-time internet connection to download MobileNetV2 base weights. Once trained, the model runs fully offline. See [Network Requirements](/frigate/network_requirements#one-time-model-downloads) for details.
+Training a custom state classification model requires an internet connection to download MobileNetV2 base weights. By default these weights are not cached in `/config/`, so they are downloaded again after the container is recreated. Once trained, the model runs fully offline. See [Network Requirements](/frigate/network_requirements#one-time-model-downloads) for details.

 :::

@@ -106,3 +106,5 @@ Output arguments are passed to FFmpeg after your camera source and control how r
 | preset-record-mjpeg              | Record - MJPEG Cameras          | Record an MJPEG stream              | Restreaming the MJPEG stream is recommended instead                                                                                                               |
 | preset-record-jpeg               | Record - JPEG Cameras           | Record a live JPEG                  | Restreaming the live JPEG is recommended instead                                                                                                                  |
 | preset-record-ubiquiti           | Record - Ubiquiti Cameras       | Record a Ubiquiti stream with audio | Handles Ubiquiti's non-standard audio format                                                                                                                      |
+
+These presets apply to the `record` output args. If [sub stream recording](/configuration/record#sub-stream-recording) is enabled, the same args are used for the `record_sub` role unless `output_args.record_sub` is set, which accepts the same presets and manual args.
@@ -6,6 +6,7 @@ title: Configuring Generative AI
 import ConfigTabs from "@site/src/components/ConfigTabs";
 import TabItem from "@theme/TabItem";
 import NavPath from "@site/src/components/NavPath";
+import FaqItem from "@site/src/components/FaqItem";

 ## Configuration

@@ -13,6 +14,18 @@ A Generative AI provider can be configured in the global config, which will make

 `genai` is a map of named providers. Each key under `genai` is a name you choose, and its value is that provider's settings:

+<ConfigTabs>
+<TabItem value="ui">
+
+1. Navigate to <NavPath path="Settings > Enrichments > Generative AI" />.
+   - Click **Add** and enter a **Provider name**. Any name of letters, numbers, hyphens, and underscores is accepted, but it cannot be changed from the UI after the provider is created.
+   - Set **Provider** to the service you are using (e.g., `ollama`)
+   - Set **Base URL**, **API key**, and **Model** as required by that provider
+   - Set **Roles** to the roles this provider should handle.
+
+</TabItem>
+<TabItem value="yaml">
+
 ```yaml
 genai:
  my_provider: # any name you like
@@ -25,9 +38,12 @@ genai:
      - chat
 ```

+</TabItem>
+</ConfigTabs>
+
 The examples on this page all use `my_provider`, but the name is arbitrary and is only used to reference the provider elsewhere in the config (for example, `semantic_search.model`).

-Each provider handles one or more **roles**: `chat`, `descriptions`, and `embeddings`. A provider handles all three by default, and each role may be assigned to exactly one provider. Define a single provider if you want it to do everything, or split the roles across several providers using the `roles` option.
+Each provider handles one or more **roles**: `chat`, `descriptions`, `embeddings`, and `transcribe`. A provider handles the first three by default; `transcribe` must always be listed explicitly, and is not available on Ollama, which has no audio input. Each role may be assigned to exactly one provider. Define a single provider if you want it to do everything, or split the roles across several providers using the `roles` option.

 If the provider you choose requires an API key, you may either directly paste it in your configuration, or store it in an environment variable prefixed with `FRIGATE_`.

@@ -43,14 +59,34 @@ Running Generative AI models on CPU is not recommended, as high inference times

 ### Recommended Local Models

-You must use a vision-capable model with Frigate. The following models are recommended for local deployment:
+#### Vision models

-| Model      | Notes                                                                                                                                                                |
-| ---------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `qwen3-vl` | Strong visual and situational understanding, enhanced ability to identify smaller objects and interactions with object.                                              |
-| `qwen3.5`  | Strong situational understanding, but missing DeepStack from qwen3-vl leading to worse performance for identifying objects in people's hand and other small details. |
-| `qwen3.6`  | Strong situational understanding, similar to qwen3-vl                                                                                                                |
-| `gemma4`   | Strong situational understanding, sometimes resorts to more vague terms like 'interacts' instead of assigning a specific action.                                     |
+You must use a vision-capable model with Frigate. The following models are recommended for local deployment of the `descriptions` and `chat` roles:
+
+| Model               | Review [frame mode](/configuration/genai/genai_review#frame-mode) | Notes                                                                                                                                                                |
+| ------------------- | --------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `qwen3-vl`          | `frames`                                                              | Strong visual and situational understanding, enhanced ability to identify smaller objects and interactions with object. Follows a sequence of frames on its own.     |
+| `qwen3.6`/`qwen3.8` | `frames`                                                              | Strong situational understanding, but missing DeepStack from qwen3-vl leading to worse performance for identifying objects in people's hand and other small details. |
+| `gemma4`            | `annotated_frames`                                                    | Strong situational understanding, sometimes resorts to more vague terms like 'interacts' instead of assigning a specific action. Loses track of activity that repeats or reverses, so it benefits from annotated frames. |
+
+#### Embedding models
+
+The `embeddings` role needs a different kind of model. Text queries are matched against the stored image embeddings, so the model must be trained to place images and text into the same vector space. A chat or description model will still return vectors when asked, but those vectors are not trained for retrieval and text searches will return poor matches with no error to indicate why.
+
+| Model                | Notes                                                                                                                                                               |
+| -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `qwen3-vl-embedding` | Multimodal embeddings for [Semantic Search](/configuration/semantic_search#genai-provider). Must be served by llama.cpp started with `--embeddings` and `--mmproj`. |
+
+#### Transcription models
+
+The `transcribe` role needs a model that accepts audio input. A text-only or vision-only model cannot serve this role. The following are recommended for local deployment of the `transcribe` role:
+
+| Model       | Notes                                                                                                                                                                                                      |
+| ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `qwen3-asr` | Dedicated speech recognition model covering 30 languages, and the better choice for transcription quality. It only transcribes, so it cannot be shared with the `descriptions` or `chat` roles.            |
+| `gemma4`    | General multimodal model that accepts audio as well as images, so one served model can cover `transcribe` alongside the other roles. Transcript quality is below `qwen3-asr`, particularly on noisy audio. |
+
+Both must be served by llama.cpp started with the matching audio `--mmproj`. llama.cpp only reports audio support when an audio projector is loaded. Without it Frigate sees the model as text-only and the `transcribe` role is unavailable in the UI. Frigate transcribes through the server's `/v1/audio/transcriptions` route, which llama.cpp serves for any audio-capable model.

 :::info

@@ -416,3 +452,82 @@ genai:

 </TabItem>
 </ConfigTabs>
+
+## FAQ
+
+<FaqItem id="how-do-i-debug-genai-issues" question="How do I debug GenAI issues?">
+
+Frigate's Generative AI features are configured and enabled separately. [Review descriptions and summaries](/configuration/genai/genai_review) live under `review.genai`, and [object descriptions](/configuration/genai/genai_objects) live under `objects.genai`. Configuring a provider on this page does not enable either feature, and enabling one does not enable the other. Decide which of the two is not working, then work through the steps below.
+
+1. Confirm a provider is available and holds the `descriptions` role.
+   - Review descriptions, review summaries, and object descriptions all use the provider that has the `descriptions` role assigned in <NavPath path="Settings > Enrichments > Generative AI > Roles" /> (`genai.<provider>.roles`).
+   - A provider is contacted the first time one of its roles is actually used. A provider holding the `embeddings` role for semantic search is initialized during startup, while a `descriptions` provider is not initialized until the first description is requested, which may be well after boot.
+   - In <NavPath path="Settings > Enrichments > Generative AI" />, use **Refresh models** next to the model field. It queries the provider for its model list and is a quick way to verify that the base URL, API key, and network path between Frigate and your provider are correct.
+
+2. Confirm the feature you expect is actually enabled.
+   - Object descriptions are disabled by default. Turn on <NavPath path="Settings > Global configuration > Objects > GenAI object config > Enable GenAI" /> (`objects.genai.enabled`), either globally or per camera. This is the most common reason custom prompts appear to be ignored while review summaries are still being generated.
+   - Review descriptions are disabled by default. Turn on <NavPath path="Settings > Global configuration > Review > GenAI config > Enable GenAI descriptions" /> (`review.genai.enabled`). Once enabled, alerts are described by default but detections are not, so a detection-only review item will never get a summary unless **Enable GenAI for detections** (`review.genai.detections`) is also on.
+
+3. If object descriptions are never requested, check the filters that skip generation.
+   - <NavPath path="Settings > Global configuration > Objects > GenAI object config > GenAI objects" /> (`objects.genai.objects`) limits generation to specific labels, and **Required zones** (`objects.genai.required_zones`) requires the object to have entered one of those zones. If either is set and does not match, Frigate skips the request silently.
+   - Thumbnails are only collected while an object is moving. Objects that go stationary early contribute fewer frames.
+   - **Use snapshots** (`objects.genai.use_snapshot`) requires snapshots to be enabled for the camera. If the snapshot cannot be read, Frigate logs `Cannot load snapshot for <id>, file not found` and no description is generated.
+   - **Send on end** (`objects.genai.send_triggers.tracked_object_end`) is on by default. If you have turned it off in favor of **Early GenAI trigger** (`objects.genai.send_triggers.after_significant_updates`), descriptions are only requested once that number of updates is reached.
+
+4. Enable debug logs to see exactly what Frigate is doing. Restart Frigate after this change. The next step also requires a restart, so turn both on at the same time to avoid restarting twice.
+
+   ```yaml
+   logger:
+     default: info
+     logs:
+       # highlight-start
+       frigate.genai: debug
+       frigate.data_processing.post.object_descriptions: debug
+       frigate.data_processing.post.review_descriptions: debug
+       # highlight-end
+   ```
+
+5. Save the exact images and prompts that were sent to your provider.
+   - Turn on **Save thumbnails** for the feature you are debugging (`review.genai.debug_save_thumbnails` or `objects.genai.debug_save_thumbnails`). Both features write to `/media/frigate/clips/genai-requests/`, and these files are admin-only.
+   - Review descriptions write `genai-requests/<review_id>/` containing the numbered frames that were sent, plus `prompt.txt` and `response.txt` with the exact prompt and the raw, unparsed model response.
+   - Review summary reports write `genai-requests/<start_ts>-<end_ts>/prompt.txt` and `response.txt`. No images are involved, since a report summarizes existing review descriptions.
+   - Object descriptions write `genai-requests/<event_id>/` containing the numbered thumbnails. The prompt for object descriptions is not written to a file, it is only visible in the debug logs from step 4.
+   - Look at the saved images before blaming the model. If the object is small, blurry, or out of frame, no prompt will fix the result. For object descriptions, consider turning on **Use snapshots** (`objects.genai.use_snapshot`) to send a higher quality image. For review items, consider setting **Review image source** (`review.genai.image_source`) to `recordings` for 480p frames instead of the lower resolution preview frames.
+
+<ConfigTabs>
+<TabItem value="ui">
+
+For review descriptions, navigate to <NavPath path="Settings > Global configuration > Review" /> and set **GenAI config > Save thumbnails** to on.
+
+For object descriptions, navigate to <NavPath path="Settings > Global configuration > Objects" />, expand **GenAI object config**, and set **Save thumbnails** to on.
+
+</TabItem>
+<TabItem value="yaml">
+
+```yaml
+review:
+  genai:
+    enabled: true
+    # highlight-next-line
+    debug_save_thumbnails: true
+
+objects:
+  genai:
+    enabled: true
+    # highlight-next-line
+    debug_save_thumbnails: true
+```
+
+</TabItem>
+</ConfigTabs>
+
+6. Verify the prompt is what you think it is.
+   - Object description prompts are the ones you control directly. A camera-level <NavPath path="Settings > Camera configuration > Objects > GenAI object config > Caption prompt" /> (`objects.genai.prompt`) overrides the global one, and an entry in **Object prompts** (`objects.genai.object_prompts`) for a label overrides both for that label. Only `{label}`, `{sub_label}`, and `{camera}` are substituted.
+   - Review description prompts are built by Frigate and request a structured JSON response, so they are not fully replaceable. The parts you control are <NavPath path="Settings > Global configuration > Review > GenAI config > Activity context prompt" /> (`review.genai.activity_context_prompt`) and **Additional concerns** (`review.genai.additional_concerns`). Keep the activity context prompt general, since overly specific rules will sway the model's threat level scoring.
+
+7. If descriptions are generated but the results are poor or inconsistent, look at the model and the context window.
+   - Empty fields, missing `shortSummary` values, or `Failed to parse review description` errors usually mean the model is not following the requested JSON schema. Smaller models struggle with structured output. Try a larger parameter size or one of the [recommended models](#recommended-local-models).
+   - Frigate calculates how many frames to send from the context size the provider reports. If your server reports a different value than it is actually running with, frames will be truncated or the request will fail. Pin the value by adding `context_size` under <NavPath path="Settings > Enrichments > Generative AI > Provider options" /> (`genai.<provider>.provider_options`), and for Ollama also confirm `options.num_ctx` there matches the context you have configured.
+   - Check **Review Description Speed** and **Object Description Speed** in <NavPath path="Health and Metrics > Enrichments" />. If inference takes tens of seconds, requests will queue behind each other and descriptions will appear to stop. For Ollama, review `OLLAMA_NUM_PARALLEL`, `OLLAMA_MAX_QUEUE`, and `OLLAMA_MAX_LOADED_MODELS` so that concurrent requests from Frigate are handled the way you expect.
+
+</FaqItem>
@@ -113,3 +113,7 @@ Many providers also have a public facing chat interface for their models. Downlo
 - OpenAI - [ChatGPT](https://chatgpt.com)
 - Gemini - [Google AI Studio](https://aistudio.google.com)
 - Ollama - [Open WebUI](https://docs.openwebui.com/)
+
+## Troubleshooting
+
+If descriptions are not being generated, or the generated descriptions are not what you expect, see [How do I debug GenAI issues?](/configuration/genai/genai_config#how-do-i-debug-genai-issues).
@@ -192,6 +192,76 @@ review:
 </TabItem>
 </ConfigTabs>

+### Frame Mode
+
+Review items are sent to the model as a sequence of still frames. Some models follow that sequence well on their own; others lose track of activity that repeats or reverses, and describe a single trip when the subject actually made several. The `frame_mode` option controls how those frames are presented.
+
+- `frames` (default): the prompt followed by the frames, exactly as earlier versions of Frigate sent them.
+- `annotated_frames`: each frame is preceded by its frame number and elapsed time, along with notes describing what the object tracker recorded at that moment, such as an object being first detected, starting to move, turning around, stopping, or no longer being detected.
+
+The notes come from tracking data rather than from the images, so they describe activity the model may not have picked up on its own. In testing with a person carrying three waste bins to the curb one at a time, `gemma4` described a single trip on every attempt with `frames`, and consistently described multiple trips with `annotated_frames`. Models that already handle these sequences well, such as the `qwen3-vl` family, gain little and should stay on `frames`.
+
+Annotated mode also caps the number of frames, since the notes already establish the order of events and extra near-duplicate frames tend to crowd out the middle of a clip. Longer review items are sampled more sparsely as a result, and typically use fewer tokens than `frames` mode for the same item.
+
+:::note
+
+Annotated mode needs tracking data for the review item. If none is available, Frigate falls back to sending plain frames for that item.
+
+:::
+
+<ConfigTabs>
+<TabItem value="ui">
+
+Navigate to <NavPath path="Settings > Global configuration > Review" />.
+
+- Set **GenAI config > Frame mode** to the desired mode (e.g., `annotated_frames`)
+
+</TabItem>
+<TabItem value="yaml">
+
+```yaml {4}
+review:
+  genai:
+    enabled: true
+    frame_mode: annotated_frames
+```
+
+</TabItem>
+</ConfigTabs>
+
+### Response Style
+
+Different models respond to the built-in prompt with very different writing styles: some produce natural narration while others sound short and mechanical. The `response_style` option selects a writing style preset that rewords the prompt's instructions for the user-facing fields (the title, short summary, and scene description). Presets replace those instructions rather than adding extra ones, so the model never receives competing style directions.
+
+Available presets:
+
+- `default`: The built-in prompt, unchanged. This already reads like a neutral security report.
+- `natural`: Plain, everyday narration with flowing sentences and sentence-style headline titles. Useful when a model's output sounds robotic.
+- `concise`: As brief as possible while still covering each significant action, with terse two-to-four word titles.
+- `detailed`: Thorough descriptions and titles that include the most identifying specifics, like colors, clothing, and carried items.
+
+Style presets only adjust how the user-facing text reads; the model's step-by-step observations and threat level scoring guidance are unaffected. Results vary by model, so it is worth comparing presets against saved debug output using `testing-scripts/genai_review_tester.py` in the Frigate repository.
+
+<ConfigTabs>
+<TabItem value="ui">
+
+Navigate to <NavPath path="Settings > Global configuration > Review" />.
+
+- Set **GenAI config > Response style** to the desired preset (e.g., `natural`)
+
+</TabItem>
+<TabItem value="yaml">
+
+```yaml {4}
+review:
+  genai:
+    enabled: true
+    response_style: natural
+```
+
+</TabItem>
+</ConfigTabs>
+
 ## Review Reports

 Along with individual review item summaries, Generative AI can also produce a single report of review items from all cameras marked "suspicious" over a specified time period (for example, a daily summary of suspicious activity while you're on vacation).
@@ -201,3 +271,7 @@ Along with individual review item summaries, Generative AI can also produce a si
 Review reports can be requested via the [API](/integrations/api/generate-review-summary-review-summarize-start-start-ts-end-end-ts-post) by sending a POST request to `/api/review/summarize/start/{start_ts}/end/{end_ts}` with Unix timestamps.

 For Home Assistant users, there is a built-in service (`frigate.review_summarize`) that makes it easy to request review reports as part of automations or scripts. This allows you to automatically generate daily summaries, vacation reports, or custom time period reports based on your specific needs.
+
+## Troubleshooting
+
+If summaries are not being generated, or the generated summaries are not what you expect, see [How do I debug GenAI issues?](/configuration/genai/genai_config#how-do-i-debug-genai-issues).
@@ -67,4 +67,6 @@ If your stream won't play, has no audio, uses excessive CPU, or otherwise misbeh

 ## Homekit Configuration

-To add camera streams to Homekit Frigate must be configured in docker to use `host` networking mode. Once that is done, you can use the go2rtc WebUI (accessed via port 1984, which is disabled by default) to export a camera to Homekit. Any changes made will automatically be saved to `/config/go2rtc_homekit.yml`.
+To export camera streams to HomeKit, Frigate must be configured in docker to use `host` networking mode. HomeKit settings are stored in `/config/go2rtc_homekit.yml` rather than in your Frigate config, and are edited through the go2rtc config editor at `http://<frigate_host>:1984/editor.html`. Pairings are saved back to that file automatically.
+
+See the [HomeKit integration docs](/integrations/homekit) for the full setup, including the video and audio requirements HomeKit places on the stream.
@@ -312,8 +312,9 @@ ffmpeg:

 :::note

-If running Frigate through Docker, you either need to run in privileged mode or
-map the `/dev/video*` devices to Frigate. With Docker Compose add:
+If running Frigate through Docker, map the relevant `/dev/video*` devices into
+the container. Running in privileged mode also works but grants far more access
+than needed. With Docker Compose add:

 ```yaml {4-5}
 services:
@@ -498,7 +499,7 @@ cameras:

 ## Synaptics

-Hardware accelerated video de-/encoding is supported on Synpatics SL-series SoC.
+Hardware accelerated video de-/encoding is supported on Synaptics SL-series SoC.

 ### Prerequisites

@@ -8,7 +8,7 @@ import TabItem from "@theme/TabItem";
 import NavPath from "@site/src/components/NavPath";
 import FaqItem from "@site/src/components/FaqItem";

-Frigate can recognize license plates on vehicles and automatically add the detected characters to the `recognized_license_plate` field or a [known](#matching) name as a `sub_label` to tracked objects of type `car` or `motorcycle`. A common use case may be to read the license plates of cars pulling into a driveway or cars passing by on a street.
+Frigate can recognize license plates on vehicles and automatically add the detected characters to the `recognized_license_plate` field or a [known](#matching) name as a `sub_label` to tracked objects of type `car`, `motorcycle`, `bus`, `truck`, `school_bus`, or `garbage_truck`, depending on which of those labels your model detects. A common use case may be to read the license plates of cars pulling into a driveway or cars passing by on a street.

 LPR works best when the license plate is clearly visible to the camera. For moving vehicles, Frigate continuously refines the recognition process, keeping the most confident result. When a vehicle becomes stationary, LPR continues to run for a short time after to attempt recognition.

@@ -24,7 +24,7 @@ When a plate is recognized, the details are:
 - Viewable in the Details pane in Review/History.
 - Viewable in the Tracked Object Details pane in Explore (sub labels and recognized license plates).
 - Filterable through the More Filters menu in Explore.
- Published via the `frigate/events` MQTT topic as a `sub_label` ([known](#matching)) or `recognized_license_plate` (unknown) for the `car` or `motorcycle` tracked object.
+- Published via the `frigate/events` MQTT topic as a `sub_label` ([known](#matching)) or `recognized_license_plate` (unknown) for the vehicle tracked object.
 - Published via the `frigate/tracked_object_update` MQTT topic with `name` (if [known](#matching)) and `plate`.

 ## Model Requirements
@@ -35,7 +35,7 @@ Users without a model that detects license plates can still run LPR. Frigate use

 :::note

-In the default mode, Frigate's LPR needs to first detect a `car` or `motorcycle` before it can recognize a license plate. If you're using a dedicated LPR camera and have a zoomed-in view where a `car` or `motorcycle` will not be detected, you can still run LPR, but the configuration parameters will differ from the default mode. See the [Dedicated LPR Cameras](#dedicated-lpr-cameras) section below.
+In the default mode, Frigate's LPR needs to first detect a vehicle before it can recognize a license plate. If you're using a dedicated LPR camera and have a zoomed-in view where a vehicle will not be detected, you can still run LPR, but the configuration parameters will differ from the default mode. See the [Dedicated LPR Cameras](#dedicated-lpr-cameras) section below.

 :::

@@ -86,7 +86,7 @@ cameras:
 </TabItem>
 </ConfigTabs>

-For non-dedicated LPR cameras, ensure that your camera is configured to detect objects of type `car` or `motorcycle`, and that a car or motorcycle is actually being detected by Frigate. Otherwise, LPR will not run.
+For non-dedicated LPR cameras, ensure that your camera is configured to detect vehicle objects, and that a vehicle is actually being detected by Frigate. Otherwise, LPR will not run. The object types that can carry a plate are defined by your model's `attributes_map`, so if your model detects other vehicle labels, you can add them there.

 Like the other real-time processors in Frigate, license plate recognition runs on the camera stream defined by the `detect` role in your config. To ensure optimal performance, select a suitable resolution for this stream in your camera's firmware that fits your specific scene and requirements.

@@ -158,7 +158,7 @@ lpr:

 Navigate to <NavPath path="Settings > Enrichments > License plate recognition" />.

- **Known plates**: Assign custom `sub_label` values to `car` and `motorcycle` objects when a recognized plate matches a known value. These labels appear in the UI, filters, and notifications. Unknown plates are still saved but are added to the `recognized_license_plate` field rather than the `sub_label`.
+- **Known plates**: Assign custom `sub_label` values to vehicle objects when a recognized plate matches a known value. These labels appear in the UI, filters, and notifications. Unknown plates are still saved but are added to the `recognized_license_plate` field rather than the `sub_label`.
 - **Match distance**: Allows for minor variations (missing/incorrect characters) when matching a detected plate to a known plate. For example, setting to `1` allows a plate `ABCDE` to match `ABCBE` or `ABCD`. This parameter will _not_ operate on known plates that are defined as regular expressions.

 </TabItem>
@@ -316,7 +316,7 @@ lpr:

 :::note

-If a camera is configured to detect `car` or `motorcycle` but you don't want Frigate to run LPR for that camera, disable LPR at the camera level:
+If a camera is configured to detect vehicles but you don't want Frigate to run LPR for that camera, disable LPR at the camera level:

 <ConfigTabs>
 <TabItem value="ui">
@@ -456,7 +456,7 @@ With this setup:
 - Snapshots will have license plate bounding boxes on them.
 - The `frigate/events` MQTT topic will publish tracked object updates.
 - Debug view will display `license_plate` bounding boxes.
- If you are using a Frigate+ model and want to submit images from your dedicated LPR camera for model training and fine-tuning, annotate both the `car` / `motorcycle` and the `license_plate` in the snapshots on the Frigate+ website, even if the car is barely visible.
+- If you are using a Frigate+ model and want to submit images from your dedicated LPR camera for model training and fine-tuning, annotate both the vehicle and the `license_plate` in the snapshots on the Frigate+ website, even if the vehicle is barely visible.

 ### Using the Secondary LPR Pipeline (Without Frigate+)

@@ -611,9 +611,9 @@ If you are still having issues detecting plates, start with a basic configuratio

 </FaqItem>

-<FaqItem id="can-i-run-lpr-without-detecting-car-or-motorcycle-objects" question={<>Can I run LPR without detecting <code>car</code> or <code>motorcycle</code> objects?</>}>
+<FaqItem id="can-i-run-lpr-without-detecting-car-or-motorcycle-objects" question={<>Can I run LPR without detecting vehicle objects?</>}>

-In normal LPR mode, Frigate requires a `car` or `motorcycle` to be detected first before recognizing a license plate. If you have a dedicated LPR camera, you can change the camera `type` to `"lpr"` to use the Dedicated LPR Camera algorithm. This comes with important caveats, though. See the [Dedicated LPR Cameras](#dedicated-lpr-cameras) section above.
+In normal LPR mode, Frigate requires a vehicle to be detected first before recognizing a license plate. If you have a dedicated LPR camera, you can change the camera `type` to `"lpr"` to use the Dedicated LPR Camera algorithm. This comes with important caveats, though. See the [Dedicated LPR Cameras](#dedicated-lpr-cameras) section above.

 </FaqItem>

@@ -697,9 +697,9 @@ lpr:
   - You may need to adjust your `detection_threshold` if your plates are not being detected.

 4. Ensure the characters on detected plates are being _recognized_.
-   - Check the **Plate recognition** inference time in Enrichment metrics (<NavPath path="System metrics > Enrichments" />). High inference times (> 100ms) could lead to poor recognition results, especially for dedicated LPR cameras where the plate crosses the frame quickly.
+   - Check the **Plate recognition** inference time in Enrichment metrics (<NavPath path="Health and Metrics > Enrichments" />). High inference times (> 100ms) could lead to poor recognition results, especially for dedicated LPR cameras where the plate crosses the frame quickly.
   - Enable `debug_save_plates` to save images of detected text on plates to the clips directory (`/media/frigate/clips/lpr`). Ensure these images are readable and the text is clear.
-   - Watch the debug view to see plates recognized in real-time. For non-dedicated LPR cameras, the `car` or `motorcycle` label will change to the recognized plate when LPR is enabled and working.
+   - Watch the debug view to see plates recognized in real-time. For non-dedicated LPR cameras, the vehicle's label will change to the recognized plate when LPR is enabled and working.
   - Adjust `recognition_threshold` settings per the suggestions [above](#advanced-configuration).

 </FaqItem>
@@ -714,13 +714,13 @@ LPR's performance impact depends on your hardware. Ensure you have at least 4GB

 The YOLOv9 license plate detector model will run (and the metric will appear) if you've enabled LPR but haven't defined `license_plate` as an object to track, either at the global or camera level.

-If you are detecting `car` or `motorcycle` on cameras where you don't want to run LPR, make sure you disable LPR it at the camera level. And if you do want to run LPR on those cameras, make sure you define `license_plate` as an object to track.
+If you are detecting vehicles on cameras where you don't want to run LPR, make sure you disable LPR it at the camera level. And if you do want to run LPR on those cameras, make sure you define `license_plate` as an object to track.

 </FaqItem>

 <FaqItem id="it-looks-like-frigate-picked-up-my-cameras-timestamp-or-overlay-text-as-the-license-plate-how-can-i-prevent-this" question="It looks like Frigate picked up my camera's timestamp or overlay text as the license plate. How can I prevent this?">

-This could happen if cars or motorcycles travel close to your camera's timestamp or overlay text. You could either move the text through your camera's firmware, or apply a mask to it in Frigate.
+This could happen if vehicles travel close to your camera's timestamp or overlay text. You could either move the text through your camera's firmware, or apply a mask to it in Frigate.

 If you are using a model that natively detects `license_plate`, add an _object mask_ of type `license_plate` and a _motion mask_ over your text.

@@ -28,6 +28,24 @@ WebRTC may use an external STUN server for NAT traversal. MSE and HLS streaming

 :::

+### Selecting a streaming technology
+
+Frigate [defaults to MSE](#why-does-frigate-prefer-mse-over-webrtc-for-live-view) for restreamed cameras by design. To use WebRTC, select it explicitly from a camera's single-camera Live view settings (the settings menu in the camera's Live view header on desktop, or the settings drawer on mobile). Three related controls work together:
+
+- **Stream**: _what_ to play. This lists the [streams you've configured](#setting-streams-for-live-ui) (for example `Main Stream` and `Sub Stream`).
+- **Force low-bandwidth mode**: a switch that always plays Frigate's built-in low-bandwidth feed (the stream assigned the `detect` role, using JSMpeg) instead of the selected stream. It works anywhere without go2rtc and is useful on slow or metered connections. While it is enabled, the stream and streaming technology selectors are disabled; your stream and technology choices are restored when you turn it off.
+- **Streaming Technology**: _how_ to play the selected stream, listing **MSE** and **WebRTC**. It is only shown for a restreamed stream.
+
+- The choices are saved **per device, per camera** in your browser's local storage.
+- **WebRTC is only selectable when it can actually work for that stream.** When it can't, the option is shown disabled with the reason inline, and a more detailed reason (the failing codecs, or why the connectivity check failed) is logged to your browser's console. Common reasons:
+  - **Not configured**: no `candidates` or `ice_servers` are set under `go2rtc.webrtc` (see [WebRTC extra configuration](#webrtc-extra-configuration)).
+  - **Could not connect**: e.g. port `8555` isn't reachable, or a STUN/TURN server is misconfigured. Frigate runs a one-time WebRTC connectivity check when the Live view opens; the option may briefly show as "checking" while it runs.
+  - **Unsupported video codec**: the stream's video codec can't be played over WebRTC in your browser, most commonly H.265/HEVC in Firefox or Edge.
+  - **Unsupported audio codec**: WebRTC needs opus or G.711 audio, so a stream whose playback audio is only AAC (without an added opus/G.711 track) can't carry audio over WebRTC. See [Audio Support](#audio-support) for how to add one.
+  - **Unsupported browser**: the browser doesn't support WebRTC.
+
+When WebRTC isn't available, Frigate automatically uses MSE (or falls back to JSMpeg), so live view keeps working regardless of the selection.
+
 ### Camera Settings Recommendations

 If you are using go2rtc, you should adjust the following settings in your camera's firmware for the best experience with Live view:
@@ -157,6 +175,17 @@ WebRTC works by creating a TCP or UDP connection on port `8555`. However, it req
        - stun:8555
  ```

+- The web UI uses the STUN and TURN servers in `ice_servers` and falls back to Google's public STUN server when none are set:
+
+  ```yaml title="config.yml"
+  go2rtc:
+    webrtc:
+      ice_servers:
+        - urls: [turn:turn.example.com:3478]
+          username: frigate
+          credential: password
+  ```
+
 - For access through Tailscale, the Frigate system's Tailscale IP must be added as a WebRTC candidate. Tailscale IPs all start with `100.`, and are reserved within the `100.64.0.0/10` CIDR block.

 - Note that some browsers may not support H.265 (HEVC). You can check your browser's current version for H.265 compatibility [here](https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#codecs-madness).
@@ -206,6 +235,8 @@ For devices that support two way talk, Frigate can be configured to use the feat
 - Ensure you access Frigate via https (may require [opening port 8971](/frigate/installation/#ports)).
 - For the Home Assistant Frigate card, [follow the docs](http://card.camera/#/usage/2-way-audio) for the correct source.

+The two-way talk control in the single-camera Live view is only enabled when WebRTC is available; if WebRTC isn't configured or can't connect, the control is shown disabled.
+
 To use the Reolink Doorbell with two way talk, you should use the [recommended Reolink configuration](/configuration/camera_specific#reolink-cameras)

 As a starting point to check compatibility for your camera, view the list of cameras supported for two-way talk on the [go2rtc repository](https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#two-way-audio). For cameras in the category `ONVIF Profile T`, you can use the [ONVIF Conformant Products Database](https://www.onvif.org/conformant-products/)'s FeatureList to check for the presence of `AudioOutput`. A camera that supports `ONVIF Profile T` _usually_ supports this, but due to inconsistent support, a camera that explicitly lists this feature may still not work. If no entry for your camera exists on the database, it is recommended not to buy it or to consult with the manufacturer's support on the feature availability.
@@ -334,7 +365,7 @@ When your browser runs into problems playing back your camera streams, it will l

 - **stalled**
  - What it means: Playback has stalled because the player has fallen too far behind live (extended buffering or no data arriving).
-  - What to try: This is usually indicative of the browser struggling to decode too many high-resolution streams at once. Try selecting a lower-bandwidth stream (substream), reduce the number of live streams open, improve the network connection, or lower the camera resolution. Also check your camera's keyframe (I-frame) interval: shorter intervals make playback start and recover faster. You can also try increasing the timeout value in the UI pane of Frigate's settings.
+  - What to try: This is usually indicative of the browser struggling to decode too many high-resolution streams at once. Try selecting a lower-bandwidth stream (substream), reduce the number of live streams open, improve the network connection, or lower the camera resolution. Also check your camera's keyframe (I-frame) interval: shorter intervals make playback start and recover faster. You can also try increasing the timeout value in <NavPath path="Settings > UI" /> .

  - Possible console messages from the player code:
    - `Buffer time (10 seconds) exceeded, browser may not be playing media correctly.`
@@ -0,0 +1,395 @@
+---
+id: non_root
+title: Running as a non-root user
+---
+
+# Running as a non-root user
+
+Frigate's services run as an unprivileged user inside the container. The main Frigate process and nginx run as `frigate`, and go2rtc runs as its own more restricted `go2rtc` user. Only the s6 init system and the certsync helper stay root.
+
+The runtime user is uid/gid `1000:1000` by default. You can change it with `PUID`/`PGID`, or bypass Frigate's user handling entirely with Docker's own `user:`.
+
+Most upgrades need nothing. Frigate aligns your volume ownership on the first boot and grants access to your hardware at startup. The sections below cover the cases that need attention: large storage volumes, network storage, and hardware the automatic grant can't reach.
+
+## Run modes
+
+| Mode                | How to enable                   | Ownership of `/config` and `/media/frigate`            | `read_only: true` |
+| ------------------- | ------------------------------- | ------------------------------------------------------ | ----------------- |
+| Default             | nothing, this is the default    | Aligned to `1000:1000` on first boot                   | Supported         |
+| `PUID`/`PGID`       | `PUID=1001`, `PGID=1001`        | Aligned to the values you set, on first boot           | Not supported     |
+| Docker-native user  | `user: "1001:1001"`             | You own it, Frigate never changes ownership            | Supported         |
+| Root (escape hatch) | `FRIGATE_RUN_AS_ROOT=true`      | Never touched                                          | Not supported     |
+| Granular root       | `FRIGATE_ROOT_SERVICES=frigate` | Aligned at boot; recordings and exports also at create | Not supported     |
+
+`PUID`/`PGID` remapping runs `usermod` at startup, which writes to `/etc/passwd`, so it can't work with a read-only root filesystem. That combination stops at startup with a message pointing here. `EXTRA_GROUPS` writes to `/etc/group` and stops the same way; use Docker's `group_add:` instead, which needs no writes inside the container. The default mode and Docker's `user:` mode both work with `read_only: true`; see [Hardened deployment](#hardened-deployment).
+
+`FRIGATE_RUN_AS_ROOT` is matched against the exact lowercase string `true`. `True`, `TRUE`, and `1` are all ignored. `FRIGATE_DEVICE_ACLS` works the same way: only the lowercase string `false` turns off the automatic device grants.
+
+### Keeping individual services root
+
+`FRIGATE_ROOT_SERVICES` takes a comma separated list of `frigate`, `go2rtc`, and `nginx`. A listed service keeps running as root, and everything else about non-root operation still applies: `PUID`/`PGID` remapping, the ownership sweep, and ownership of the files those services create.
+
+There are two reasons to use it:
+
+- Your detector hardware won't work as an unprivileged user, even after reading [Hardware device access](#hardware-device-access). `FRIGATE_ROOT_SERVICES=frigate` keeps the main process and its detectors as root while nginx and go2rtc stay unprivileged.
+- You want everything to run as root but still want your files owned by `PUID`/`PGID` instead of root. `FRIGATE_ROOT_SERVICES=frigate,go2rtc,nginx` does that.
+
+Try the device grants and `EXTRA_GROUPS` first. The `frigate` service runs the API and every ffmpeg process that decodes your camera streams, so listing it puts those back on root as well, not just your detectors.
+
+A listed service also stops honoring a [custom ffmpeg or go2rtc build](/configuration/advanced/system#custom-dependencies) kept in `/config`, since that directory stays owned by the unprivileged user and a binary there would run as root. `FRIGATE_RUN_AS_ROOT=true` has no such restriction.
+
+Recordings and exports are owned by `PUID`/`PGID` as soon as they're written, even by a root service. Snapshots, thumbnails, and other files under `clips/` are corrected on each restart, so they can show as root-owned from the host until then. A listed service also keeps root's home directory, so library caches go to the container layer instead of `/config`. The same applies to the [detector runtimes](/frigate/network_requirements#detector-runtimes) Frigate installs at first start (Hailo, MemryX, AXEngine): a root `frigate` service installs them into `/root/.local`, which is lost when the container is recreated, and never loads a copy left behind in `/config/.local`.
+
+Listing all three services is not the same as `FRIGATE_RUN_AS_ROOT=true`. The escape hatch never touches ownership; the list keeps the ownership handling active. A few more details:
+
+- An unknown name in the list stops the container at startup, rather than silently leaving a service unprivileged.
+- Changing the list runs the full ownership sweep once on the next boot.
+- If both are set, `FRIGATE_RUN_AS_ROOT=true` wins and the list is ignored.
+- With Docker's `user:`, the list does nothing, since the container never has root to keep.
+
+## Migrating an existing install
+
+Volumes from earlier versions of Frigate are owned by root, so ownership has to be aligned with the runtime user once. This happens automatically on the first boot after upgrading.
+
+On large recordings volumes, do it from the host beforehand instead. The boot sweep runs before any service starts, so a multi-terabyte `/media/frigate` can hold the container in startup long enough for Docker's healthcheck to mark it unhealthy, and orchestrators that watch health will restart it mid-sweep. If you'd rather not run the script, raise the healthcheck start period instead (`--start-period=1800s`, or `start_period: 1800s` under `healthcheck:` in compose).
+
+Grab [`fix-permissions.sh`](https://github.com/blakeblackshear/frigate/blob/dev/docker/migration/fix-permissions.sh) from the Frigate repo and dry run it first:
+
+```bash
+./fix-permissions.sh --dry-run /path/to/your/config /path/to/your/storage
+```
+
+That reports how many entries would change and touches nothing. When it looks right, run it without `--dry-run`:
+
+```bash
+./fix-permissions.sh /path/to/your/config /path/to/your/storage
+```
+
+Pass `PUID` and `PGID` as the third and fourth arguments if you're not using the default `1000:1000`. The script wraps the same helper the container uses, so the result is identical either way. Override the image it pulls with `FRIGATE_IMAGE=...` if you're not on `stable`.
+
+Both the script and the boot sweep report progress, so you can tell a slow sweep from a stuck one:
+
+```
+[INFO] fix-ownership: scanning /media/frigate for ownership mismatches; this may take a while on large filesystems
+[WARN] fix-ownership: adjusting ownership of 4823941 entries under /media/frigate
+[INFO] fix-ownership: /media/frigate 5% (241197/4823941 entries)
+[INFO] fix-ownership: /media/frigate 10% (482394/4823941 entries)
+[INFO] fix-ownership: finished /media/frigate in 12m 4s
+```
+
+The scan has no percentage because the total isn't known until it finishes. Watch the boot sweep with `docker logs -f frigate`.
+
+Once the volumes are aligned, start Frigate normally. A file at `/config/.permissions_version` records what was done, so later boots skip the sweep unless you change `PUID`/`PGID`.
+
+If something under your volumes can't be chowned, a read-only btrfs snapshot directory for example, the sweep warns and names the path and doesn't record the migration as finished. It retries on the next boot instead. Either move those paths outside `/media/frigate` or expect the scan to repeat.
+
+### Network storage
+
+Recordings on a NAS behave differently, so check what you have before migrating:
+
+```bash
+findmnt -T /path/to/your/storage -o TARGET,FSTYPE,OPTIONS
+```
+
+**SMB and CIFS** don't store per-file ownership at all. It's synthesized from the mount options, so a per-file `chown` fails and isn't needed. Mount the share as the uid and gid Frigate runs as, and every file already looks correct to the sweep:
+
+```
+//nas/frigate /media/frigate cifs credentials=/root/.smb,uid=1000,gid=1000,file_mode=0664,dir_mode=0775 0 0
+```
+
+**NFS** exports default to `root_squash` on most servers, which maps the container's root to `nobody`. The chown then fails, you get `[WARN] fix-ownership: some entries under /media/frigate could not be updated`, and since the sweep didn't finish it doesn't record the migration, so it retries on every boot.
+
+The best fix is to not chown over NFS at all. Do it on the server, where there's no squash and no network round trip per file:
+
+```bash
+# on the NAS itself, against the exported directory
+chown -R 1000:1000 /export/frigate
+```
+
+Frigate's sweep then finds nothing to change and records the migration normally. If you can't get a shell on the server, you can export temporarily with `no_root_squash`, migrate, and put it back, or leave ownership alone and set `PUID`/`PGID` to whichever uid already owns the files.
+
+Either way the uid has to mean the same thing on both machines. NFS sends numeric uids, so container uid 1000 is uid 1000 on the server no matter what the usernames are.
+
+Expect the first boot to be slow even when nothing needs changing, because checking ownership costs a round trip per file. That's a one-time cost. **If the sweep runs on every boot rather than once, ownership isn't actually being applied**, and the warning above will say so.
+
+Keep `/config` on local storage either way. Frigate's database is SQLite and network shares handle its locking poorly. That's a long-standing recommendation, not something running non-root introduces.
+
+## Rolling back
+
+Set `FRIGATE_RUN_AS_ROOT=true` and restart. Everything runs as root again, exactly as it did before. This is the fastest way to get a broken install running while you sort out a device permission problem.
+
+The escape hatch never changes ownership, and it clears the record of the last sweep on startup, so switching back to non-root later corrects whatever root created in the meantime. Toggling in either direction is safe.
+
+## Hardware device access
+
+Frigate grants the runtime user access to your devices at startup. Pass your hardware with `--device` (or `devices:` in compose) and detection and hardware acceleration work with no group or udev setup on the host.
+
+The grant covers the common accelerator and camera nodes: GPU render nodes, Intel/AMD NPUs (`/dev/accel`), Coral, Hailo, Rockchip, Jetson, `/dev/video*`, and the USB bus. For hardware it misses, add your own paths with `DEVICE_ACL_PATHS`, a comma separated list of globs:
+
+```yaml
+environment:
+  DEVICE_ACL_PATHS: "/dev/mydev*"
+```
+
+Set `FRIGATE_DEVICE_ACLS=false` if you manage device permissions yourself and want Frigate to leave them alone.
+
+Frigate grants access by adding an ACL entry for the runtime users. The device's owner and mode are unchanged, and nothing is made world accessible. One thing to know: `--device` nodes belong to the container, but a bind mounted `/dev/bus/usb` (the usual Coral USB setup) shares the host's device nodes, so the entry is visible on the host until udev recreates the node.
+
+### Manual setup
+
+You only need this for hardware the automatic grant can't reach, or for Docker's `user:` mode, where there's no root startup to do the granting.
+
+Your accelerator most likely worked in older versions because Frigate ran as root. Device nodes are usually owned by `root:root`, and root either matches the group or skips the check entirely. The runtime user does neither, so a device that worked before can become unreadable with no change to your Frigate config.
+
+#### Read what your device requires
+
+Find the node and look at its owner, group, and mode:
+
+```bash
+ls -ln /dev/dri/renderD128
+crw-rw---- 1 0 105 226, 128 Jul  5 10:12 /dev/dri/renderD128
+#          ^ ^  ^
+#          | |  group GID 105
+#          | owner UID 0 (root)
+#          mode: owner rw, group rw, other none
+```
+
+Then work out which of the three permission sets applies to the runtime user. It isn't the owner, since that's root, so it gets the group bits if it belongs to that GID and otherwise falls through to "other". In the example above "other" is empty, so without membership in group 105 the runtime user can't open the node.
+
+Watch for a node that looks permissive but isn't. A USB Coral defaults to this:
+
+```bash
+ls -ln /dev/bus/usb/004/003
+crw-rw-r-- 1 0 0 189, 386 Jul  5 10:12 /dev/bus/usb/004/003
+```
+
+The group is `0`, so "other" applies to the runtime user, and "other" here is read only. `libedgetpu` needs to write to the node, so detection fails with `No EdgeTPU was detected` as though no Coral were attached. Read access alone isn't enough for most accelerators.
+
+#### Grant access
+
+Give the runtime user the GID with `EXTRA_GROUPS`, a comma separated list of numeric host GIDs. They're added to both the `frigate` and `go2rtc` users, which matters because go2rtc needs its own render and video access for hardware accelerated restreams.
+
+```yaml
+environment:
+  EXTRA_GROUPS: "105,44" # host render and video GIDs
+```
+
+Use numeric GIDs from the host, not names. Group names don't have to match between the host and the container, and the kernel only checks the number. If the GID doesn't exist in the image, Frigate creates a placeholder group for it.
+
+Two things that look like they should work but don't:
+
+- Docker's `group_add` has no effect in the default or `PUID` modes. Frigate rebuilds the supplementary group list from `/etc/group` when it drops privileges, which discards what Docker passed in. It is the right tool with Docker's `user:`, where no privilege drop happens and `EXTRA_GROUPS` does nothing.
+- `privileged: true` doesn't help. It grants capabilities to root, and the runtime user isn't root, so the file permissions on the node still apply.
+
+If the node's group is `root` or the mode denies the group, no `EXTRA_GROUPS` value will help. You need a udev rule first.
+
+#### Verify access
+
+Check the group landed, then check the runtime user can open the node. Test for write, not just read:
+
+```bash
+docker exec frigate id frigate
+docker exec frigate /command/s6-setuidgid frigate sh -c 'test -w /dev/dri/renderD128 && echo ok'
+docker exec frigate /command/s6-setuidgid go2rtc  sh -c 'test -w /dev/dri/renderD128 && echo ok'
+```
+
+A permission check is only a proxy for the driver working. These exercise the real libraries as the runtime user:
+
+```bash
+docker exec frigate /command/s6-setuidgid frigate vainfo
+docker exec frigate /command/s6-setuidgid frigate python3 -c "import openvino as ov; print(ov.Core().available_devices)"
+```
+
+`vainfo` should reach `va_openDriver() returns 0` and list profiles. Complaints about `XDG_RUNTIME_DIR` or an X server above that are normal. OpenVINO should list `GPU`; if it returns only `CPU`, detection has fallen back and inference will be much slower without an error in the log.
+
+To tell a permissions problem from anything else, start the container once with `FRIGATE_RUN_AS_ROOT=true`. If the device works as root and not otherwise, it's node permissions and a udev rule is the fix. If it's missing either way, the problem is your device mapping or the host, and isn't related to running non-root.
+
+#### udev rules by device
+
+Rules go in `/etc/udev/rules.d/` on the host and take effect after:
+
+```bash
+sudo udevadm control --reload-rules && sudo udevadm trigger
+```
+
+A device that's already connected sometimes keeps its original ownership through a trigger. If `ls -ln` doesn't show the new group, replug it, or reboot for a built-in device.
+
+**Coral USB** needs two rules, because the device re-enumerates after loading firmware. It appears as Global Unichip `1a6e` before and Google `18d1` after, with a different node each time. A rule covering only `1a6e` gives you a Coral that starts up once and then disappears mid-run.
+
+```
+SUBSYSTEM=="usb", ATTRS{idVendor}=="1a6e", GROUP="plugdev", MODE="0664"
+SUBSYSTEM=="usb", ATTRS{idVendor}=="18d1", GROUP="plugdev", MODE="0664"
+```
+
+Map the whole `/dev/bus/usb` rather than a single node, for the same reason. Most hosts put `plugdev` at GID 46 and the image agrees, so a USB Coral often needs no `EXTRA_GROUPS` entry. Confirm with `getent group plugdev` and add the number if your host differs.
+
+**Coral PCIe** is often `crw------- root root`, which only root can open:
+
+```
+SUBSYSTEM=="apex", MODE="0660", GROUP="apex"
+```
+
+Create the group with `sudo groupadd -f apex`, then add its GID to `EXTRA_GROUPS`.
+
+**Hailo** works the same way. Grant `/dev/hailo0` a group and add that GID:
+
+```
+SUBSYSTEM=="hailo_chardev", MODE="0660", GROUP="hailo"
+```
+
+**Intel and AMD GPUs** usually need nothing beyond `EXTRA_GROUPS`, since most distributions ship a `render` group that owns `/dev/dri/renderD128`. The GID often differs between the host and the image, so pass the host's number rather than assuming the name resolves. Debian based images have no `render` group at all.
+
+#### Quick reference
+
+What each device needs when you're setting it up by hand. The automatic grant covers most of these already, so start here only if it didn't.
+
+| Hardware                  | Device(s)                                                   | What non-root needs                                                                                              |
+| ------------------------- | ----------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- |
+| Intel/AMD GPU (VAAPI/QSV) | `/dev/dri/renderD128`                                       | Host render GID in `EXTRA_GROUPS`, from `getent group render`                                                    |
+| Intel/AMD NPU             | `/dev/accel`                                                | udev rule granting a group, then that GID in `EXTRA_GROUPS`                                                      |
+| Coral USB                 | `/dev/bus/usb`                                              | udev rules for both `1a6e` and `18d1`; usually already covered by `plugdev` 46                                   |
+| Coral PCIe                | `/dev/apex_0`                                               | udev rule granting a group, then that GID in `EXTRA_GROUPS`                                                      |
+| Hailo                     | `/dev/hailo0`                                               | udev rule granting a group, then that GID in `EXTRA_GROUPS`                                                      |
+| NVIDIA                    | nvidia runtime                                              | Nothing, works with the nvidia-container-toolkit defaults                                                        |
+| AMD ROCm                  | `/dev/kfd`, `/dev/dri`                                      | Host `video` and `render` GIDs in `EXTRA_GROUPS`                                                                 |
+| Raspberry Pi              | `/dev/video11`                                              | Host `video` GID in `EXTRA_GROUPS`                                                                               |
+| Rockchip                  | `/dev/dri`, `/dev/dma_heap`, `/dev/rga`, `/dev/mpp_service` | Commonly `root:root` `0600`, so all four need udev rules. If you can't grant all four, use `FRIGATE_RUN_AS_ROOT` |
+| Axera (AXCL)              | `/dev/ax_*` per the AXCL driver docs                        | Unverified. Check node ownership on your hardware before assuming this works                                     |
+| Synaptics SL1680          | per the Synaptics docs                                      | Unverified                                                                                                       |
+| MemryX                    | per the MemryX docs                                         | Still requires `privileged: true`, which means root. Out of scope for non-root operation                         |
+| Nvidia Jetson             | nvidia runtime plus Jetson nodes                            | Unverified. The nvidia runtime handles mapping, but check `/dev/nvhost-*` ownership on your board                |
+| VeriSilicon NPU (Teflon)  | per the driver, commonly `/dev/galcore`                     | Unverified. Check node ownership on your hardware before assuming this works                                     |
+| CPU detector              | none                                                        | Nothing, no device is opened                                                                                     |
+| ZMQ detector              | none                                                        | Nothing, inference happens over a socket                                                                         |
+| Apple Silicon             | none                                                        | Nothing, the NPU client runs on the host and Frigate reaches it over the network                                 |
+
+## Hardened deployment
+
+A read-only root filesystem means the container can't modify itself, only the volumes you give it. It works in the default mode and under Docker's `user:`, but not with `PUID`/`PGID` or `EXTRA_GROUPS`, which both need to write to `/etc`.
+
+Start with the default mode. It keeps go2rtc on its own restricted user and still grants your hardware automatically, at the cost of a short root startup that finishes before any service runs.
+
+```yaml
+services:
+  frigate:
+    container_name: frigate
+    image: ghcr.io/blakeblackshear/frigate:stable
+    restart: unless-stopped
+    stop_grace_period: 30s
+    read_only: true
+    security_opt:
+      - no-new-privileges:true
+    shm_size: "512mb" # size for your cameras, see the shm-size calculation
+    devices:
+      - /dev/dri/renderD128:/dev/dri/renderD128 # your hardware, granted at startup
+    volumes:
+      - /etc/localtime:/etc/localtime:ro
+      - /path/to/your/config:/config
+      - /path/to/your/storage:/media/frigate
+    tmpfs:
+      - /tmp:size=256m
+      - /tmp/cache:size=1000000000 # recording segments, sized as before
+      - /run:exec,nosuid,nodev,mode=0755,size=16m
+    ports:
+      - "8971:8971"
+      - "8554:8554" # RTSP feeds
+      - "8555:8555/tcp" # WebRTC over tcp
+      - "8555:8555/udp" # WebRTC over udp
+```
+
+`/run` has to allow `exec`. With a read-only root filesystem s6 copies its service scripts into `/run` and runs them from there, and tmpfs mounts default to `noexec`. The equivalent for `docker run` is `--tmpfs /run:exec,nosuid,nodev,mode=0755`. Spelling out `nosuid` and `nodev` matters: passing any tmpfs options replaces Docker's defaults instead of adjusting them, so asking for `exec` alone would drop those two as well.
+
+Size `/tmp` deliberately. It now carries nginx's config copy and its five proxy temp directories as well as the recording cache. Keeping `/tmp/cache` as its own nested tmpfs, as above, leaves your existing [cache sizing](/frigate/installation#storage) untouched and adds a small allowance for nginx. If you'd rather use one tmpfs over all of `/tmp`, size it as your cache budget plus roughly 50MB, or recordings begin failing once the cache fills.
+
+The self signed certificate is written to `/config/tls`, which stays writable. Certificates you mount at `/etc/letsencrypt/live/frigate` work unchanged and still take precedence.
+
+[Detector runtimes](/frigate/network_requirements#detector-runtimes) that Frigate installs at first start (Hailo, MemryX, AXEngine) are staged in `/tmp` and installed into `/config/.local`, so they work with a read-only root filesystem in the default mode and under `user:`. A root `frigate` service installs into `/root/.local` instead, which a read-only root filesystem prevents; either leave `frigate` out of `FRIGATE_ROOT_SERVICES` or drop `read_only`.
+
+Soak a hardened deployment for 24 hours against real cameras before relying on it. A read-only root filesystem turns an occasional write into a failure that startup won't reveal.
+
+### Never starting as root
+
+To remove root from the container entirely, add Docker's `user:`:
+
+```yaml
+    user: "1000:1000" # NOT compatible with PUID/PGID, see the run modes table
+    tmpfs:
+      - /tmp:size=256m
+      - /tmp/cache:size=1000000000
+      - /run:exec,nosuid,nodev,mode=0755,uid=1000,gid=1000,size=16m # uid must match user:
+```
+
+`/run` has to be owned by that uid as well. s6 writes its runtime state there before anything else starts, and with no root in the container a root-owned `/run` stops it during init with `cannot create /run/test of writability`. Keep `uid` and `gid` in the tmpfs options matching `user:`, and don't carry that pair back into the default mode, where a root-owned `/run` is what keeps the unprivileged services out of s6's runtime state.
+
+This only bites once root is genuinely gone. s6's init helper is setuid, so `user:` on its own still lets init regain root and correct `/run` itself. The `no-new-privileges:true` above is what blocks that, which is also what makes the `/run` ownership mandatory. Dropping it would hide the problem by handing init root again.
+
+Two things change, and the first one will break a working install if you skip it. The startup device grants can't run, because there is no root left to run them, so every device you pass stops working until you grant that uid access yourself with `group_add:` or a udev rule; see [Manual setup](#manual-setup). Expect this to surface as a driver error rather than a permission error, like `No VA display found` from VAAPI. And every service then runs as that one uid, so go2rtc no longer gets its own restricted user. `/config` and `/media/frigate` have to be owned by that uid already, since Frigate never adjusts ownership in this mode. Switching an existing install over also leaves `/config/go2rtc_homekit.yml` owned by the go2rtc user, which this mode can't write; `chown` it to your uid or HomeKit pairing changes stop persisting. Frigate warns and starts either way.
+
+This mode can also take `cap_drop: [ALL]`, which the default mode cannot: starting as root needs `CAP_CHOWN` for the ownership sweep, `CAP_SETUID` and `CAP_SETGID` to drop to the runtime user, and `CAP_FOWNER` for the device grants.
+
+### Per-variant exceptions
+
+- **Rockchip** needs `- /sys/:/sys/:ro` alongside its device nodes, in addition to everything above.
+- **MemryX** and **QNAP Container Station** still require `privileged: true` per their own documentation, which gives back most of what this layout removes. MemryX also downloads its models to `/memryx_models` on the root filesystem, so it can't run read-only regardless. Its SDK is installed into `/config/.local` like the other detector runtimes.
+
+## Network isolation
+
+Everything above limits what a compromised container can do to the host. It doesn't limit what your cameras can do to your network. Camera firmware is closed source, rarely patched, and not something you can audit, and none of it needs internet access for Frigate to work.
+
+Put the cameras on their own VLAN or subnet, give the Frigate host a route into it, and deny that VLAN any route out. Frigate reaches in to pull streams, the cameras reach nothing. A second NIC on the Frigate host is the simplest version of this, and a tagged VLAN on the NIC you already have works just as well.
+
+Here's the deny as nftables on the router, with cameras on `vlan20` and the Frigate host at `192.168.10.5`:
+
+```
+table inet cameras {
+  chain forward {
+    type filter hook forward priority filter; policy accept;
+
+    ct state established,related accept
+    iifname "vlan20" ip daddr 192.168.10.5 accept
+    iifname "vlan20" drop
+  }
+}
+```
+
+It's in its own table so it can sit alongside an existing ruleset without touching it. Streams keep working because Frigate opens those connections and the return traffic is `established`. Cameras can still reach each other on their own VLAN, since that traffic never reaches the router, so use client isolation on the switch if that matters to you.
+
+Two things break when you do this. The manufacturer's phone app stops working, which is the point, and camera clocks drift, because most of them set their time over NTP and are bad at it. Point them at an NTP server on your own network rather than opening the VLAN back up, or their timestamps and Frigate's will disagree.
+
+Frigate itself needs some outbound access, though nearly all of it is optional. The startup version check is the only piece that's on by default, and `telemetry.version_check: false` turns it off. Everything else (model downloads for the enrichment features, push notifications, Frigate+, and cloud GenAI providers) only reaches out once you enable that feature. See [Network Requirements](/frigate/network_requirements) for the full list and how to run fully offline.
+
+For containers that only talk to each other, an internal compose network gets you the same isolation without involving the router:
+
+```yaml
+services:
+  frigate:
+    networks: [default, iot]
+    # the rest of your frigate service
+  mosquitto:
+    image: eclipse-mosquitto
+    networks: [iot]
+
+networks:
+  iot:
+    internal: true
+```
+
+`internal: true` gives that network no route off the host, so the broker isn't reachable from anywhere else on your LAN. Frigate sits on both networks and keeps its normal outbound path.
+
+One Docker specific trap: published ports are inserted ahead of the host firewall, so `ufw deny 8971` doesn't do what it looks like it does. Bind the port to the interface you want instead, like `127.0.0.1:8971:8971` for a reverse proxy on the same host, or your LAN address for everything else.
+
+## Known limitations
+
+`telemetry.stats.network_bandwidth` uses nethogs, which needs `CAP_NET_ADMIN` and `CAP_NET_RAW` and therefore root. The stat is turned off automatically when Frigate isn't running as root, with one warning in the log. Use `FRIGATE_ROOT_SERVICES=frigate` (or `FRIGATE_RUN_AS_ROOT=true`) if you need it.
+
+go2rtc's ffmpeg processes no longer appear in Intel GPU stats. Frigate reads per-process GPU usage from `/proc/<pid>/fdinfo`, which the kernel won't let one user read for another user's processes, so anything go2rtc spawns is invisible to it. Overall GPU utilization is unaffected.
+
+If you mount your own TLS certificate at `/etc/letsencrypt/live/frigate`, the private key has to be readable by the runtime user, which runs nginx. Frigate hands the key to that user at startup if the mount is writable; on a read-only mount, make the key readable by uid 1000 (or your `PUID`) yourself.
+
+If you're debugging nginx, run the config check as the runtime user with stdout discarded:
+
+```bash
+docker exec frigate /command/s6-setuidgid frigate bash -c 'nginx -t -c /tmp/nginx/conf/nginx.conf >/dev/null'
+```
+
+Running `nginx -t` as root hands nginx's runtime directories to root as a side effect, which breaks the running workers until the service restarts, and the config's `/dev/stdout` logs can't be reopened through a root-owned `docker exec` pipe. The results print on stderr either way.
@@ -6,6 +6,7 @@ title: Notifications
 import ConfigTabs from "@site/src/components/ConfigTabs";
 import TabItem from "@theme/TabItem";
 import NavPath from "@site/src/components/NavPath";
+import FaqItem from "@site/src/components/FaqItem";

 # Notifications

@@ -21,7 +22,7 @@ Push notifications require internet access from the Frigate server to the browse

 In order to use notifications the following requirements must be met:

- Frigate must be accessed via a secure `https` connection ([see the authorization docs](/configuration/authentication)).
+- Frigate must be accessed via a secure `https` connection while signed in as a Frigate user ([see the authorization docs](/configuration/authentication)).
 - A supported browser must be used. Currently Chrome, Firefox, and Safari are known to be supported.
 - In order for notifications to be usable externally, Frigate must be accessible externally.
 - For iOS devices, some users have also indicated that the Notifications switch needs to be enabled in iOS Settings --> Apps --> Safari --> Advanced --> Features.
@@ -85,7 +86,13 @@ cameras:

 ### Registration

-Once notifications are enabled, press the `Register for Notifications` button on all devices that you would like to receive notifications on. This will register the background worker. After this Frigate must be restarted and then notifications will begin to be sent.
+Once notifications are enabled, press the `Register This Device` button on all devices that you would like to receive notifications on. This will register the background worker. After this Frigate must be restarted and then notifications will begin to be sent.
+
+:::warning
+
+Each registration is attached to the Frigate user account you are signed in as, so you must register over a secure connection to the authenticated port (`8971`). Reverse proxies and tunnels should point at port `8971`.
+
+:::

 ## Supported Notifications

@@ -104,3 +111,62 @@ Different platforms handle notifications differently, some settings changes may
 ### Android

 Most Android phones have battery optimization settings. To get reliable Notification delivery the browser (Chrome, Firefox) should have battery optimizations disabled. If Frigate is running as a PWA then the Frigate app should have battery optimizations disabled as well.
+
+## Notifications FAQ
+
+<FaqItem id="how-do-i-debug-notifications-issues" question="How do I debug notifications issues?">
+
+Push notifications involve Frigate, your browser, and your browser vendor's push service, so it helps to work from the server outward.
+
+1. Enable debug logs for the push client by adding `frigate.comms.webpush: debug` to your `logger` configuration. Restart Frigate after this change.
+
+   ```yaml
+   logger:
+     default: info
+     logs:
+       # highlight-next-line
+       frigate.comms.webpush: debug
+   ```
+
+   These logs show exactly where a notification stopped, including:
+   - `Email must be provided for push notifications to be sent` means the global `email` field is empty and nothing will ever be sent.
+   - `Sending test notification` and `Sending push notification for <camera>, review ID <id>` mean Frigate handed the message off to the push service.
+   - `Skipping notification for <camera> - in global cooldown period` (or `camera-specific cooldown period`) means your [cooldown](#configuration) values suppressed it.
+   - `Notifications for <camera> are currently suspended` means notifications were suspended from <NavPath path="Settings > Notifications" /> or MQTT.
+   - `Notification endpoint expired for <user>, received 410` means that device's subscription is no longer valid and it must be re-registered.
+   - `Failed to send notification to <user> :: <status>` means the push service rejected the message. A `401` or `403` usually points at a VAPID or `email` problem, and a `5xx` is a problem on the push service's end.
+   - If you see no messages at all when an alert occurs, the notification was never queued. Confirm an actual **alert** was created (notifications are not sent for detections), and that notifications are enabled both globally and for that camera.
+
+2. Verify the basics that most reports come down to:
+   - Frigate must be reached over `https` with a certificate your device trusts. Browsers silently refuse to register a service worker otherwise, and a self-signed certificate that is not installed as trusted on the device will fail.
+   - On iOS, notifications only work when Frigate has been installed to the Home Screen via **Share > Add to Home Screen** and opened from that icon. Safari and Chrome tabs cannot receive web push on iOS.
+   - Each device must be registered individually, and Frigate must be restarted after registering before anything can be sent, including test notifications.
+   - The Frigate server needs outbound internet access to the browser vendor's push service. See [Network Requirements](/frigate/network_requirements#push-notifications).
+
+3. Test from the UI. Use the `Send a test notification` button in <NavPath path="Settings > Notifications" />. If the log shows `Sending test notification` but nothing arrives on the device, the problem is between the push service and your device rather than in Frigate.
+
+4. Check the browser side on the device that is not receiving notifications:
+   - Confirm the site's notification permission is set to **Allow** in your browser or OS settings, and that a focus/do not disturb mode is not hiding them.
+   - In desktop browsers, open Developer Tools > Application > Service Workers and confirm `notifications-worker.js` is registered and activated. Unregistering it and registering the device again will rebuild a broken subscription.
+   - Check the browser console and your reverse proxy logs for failures loading `/notifications-worker.js` or errors on `/api/notifications/register`.
+
+</FaqItem>
+
+<FaqItem id="why-did-notifications-stop-arriving-after-working-for-a-while" question="Why did notifications stop arriving after working for a while?">
+
+Push subscriptions are issued by the browser vendor and can be revoked, most often after a browser update, after clearing site data, or when a device has been offline for an extended period. When this happens the device still appears registered in Frigate, but the push service rejects the message. The debug logs will show `Notification endpoint expired` with a `404` or `410` status.
+
+Unregister and re-register the affected device from <NavPath path="Settings > Notifications" />, then restart Frigate.
+
+</FaqItem>
+
+<FaqItem id="why-am-i-not-getting-notifications-for-one-specific-camera" question="Why am I not getting notifications for one specific camera?">
+
+Work through these in order:
+
+- Notifications are only sent for **alerts**. If the camera is producing detections instead, adjust the camera's `review > alerts > labels` so the objects you care about are classified as alerts.
+- Confirm notifications are enabled for that camera in <NavPath path="Settings > Camera configuration > Notifications" />.
+- Check the camera's `cooldown` value, and remember that the global cooldown applies across all cameras. A busy camera can consume the global cooldown and suppress a quieter one.
+- If [authentication](/configuration/authentication) is enabled with roles, users only receive notifications for the cameras their role grants access to.
+
+</FaqItem>
@@ -22,9 +22,9 @@ Frigate supports multiple different detectors that work on different types of ha
 **Most Hardware**

 - [Coral EdgeTPU](#edge-tpu-detector): The Google Coral EdgeTPU is available in USB, Mini PCIe, and m.2 formats allowing for a wide range of compatibility with devices.
- [Hailo](#hailo-8): The Hailo8 and Hailo8L AI Acceleration module is available in m.2 format with a HAT for RPi devices, offering a wide range of compatibility with devices.
+- [Hailo](#hailo): The Hailo-8, Hailo-8L and Hailo-8R AI Acceleration modules are available in m.2 format with a HAT for RPi devices, offering a wide range of compatibility with devices.
 - <CommunityBadge /> [MemryX](#memryx-mx3): The MX3 Acceleration module is available in m.2 format, offering broad compatibility across various platforms.
- <CommunityBadge /> [DeGirum](#degirum): Service for using hardware devices in the cloud or locally. Hardware and models provided on the cloud on [their website](https://hub.degirum.com).
+- <CommunityBadge /> [DEEPX](#deepx-npu): The DEEPX NPU is available in m.2 format and as a HAT+ for the Raspberry Pi 5, offering broad compatibility across various platforms.

 **AMD**

@@ -69,19 +69,73 @@ Frigate supports multiple different detectors that work on different types of ha

 :::note

-Multiple detectors can not be mixed for object detection (ex: OpenVINO and Coral EdgeTPU can not be used for object detection at the same time).
+A single model can not be spread across different detector types (ex: OpenVINO and Coral EdgeTPU can not run the same model at the same time). Configuring more than one model, each on its own detector type, is supported.

 This does not affect using hardware for accelerating other tasks such as [semantic search](./semantic_search.md)

 :::

+### Configuring models and hardware
+
+Object detection is configured with a `models` list. Each entry describes one model and the hardware it runs on:
+
+```yaml
+models:
+  - devices:
+      - openvino:GPU
+    path: /config/model_cache/yolov9-s.onnx
+    model_type: yolo-generic
+    width: 320
+    height: 320
+```
+
+Each entry in `devices` is a detector type, optionally followed by a colon and a device for that detector, such as `edgetpu:pci:0`, `openvino:NPU`, or `tensorrt:0`. The per-detector sections below document the device values each one accepts. Listing several devices runs the model on all of them, and listing the **same** device more than once runs additional inference processes against it, which can improve throughput on hardware that keeps up with more than one stream:
+
+```yaml
+models:
+  - devices:
+      - openvino:GPU
+      - openvino:GPU
+```
+
+Coral EdgeTPU and MemryX accelerators can only be opened by one process, so those devices can not be repeated.
+
+### Running more than one model
+
+Cameras can be split across models by scene, which is useful when indoor and outdoor cameras benefit from differently trained models. Each model declares the `scene` it is for, and each camera picks one with `detect -> scene`:
+
+```yaml
+models:
+  - scene: outdoor
+    path: plus://your-outdoor-model
+    devices:
+      - edgetpu:pci:0
+  - scene: indoor
+    path: /config/model_cache/indoor.onnx
+    model_type: yolo-generic
+    devices:
+      - openvino:GPU
+
+cameras:
+  driveway:
+    detect:
+      scene: outdoor
+    ...
+  hallway:
+    detect:
+      scene: indoor
+    ...
+```
+
+Available scenes are `all`, `indoor`, `outdoor`, `indoor_thermal`, and `outdoor_thermal`. A model with a scene of `all` is used by every camera that does not set one, and `all` is the default when a model does not declare a scene. Changing a camera's scene requires a restart.
+
 ### Choosing a model size

 Along with picking a detector for your hardware, you will choose a model's **input resolution** (such as `320x320` or `640x640`) and, for model families like YOLOv9, a **variant size** (`tiny`, `small`, etc.). Both affect the balance between accuracy and the inference time your hardware can sustain.

 **Resolution (320x320 vs 640x640):** Frigate is optimized for `320x320` models, and `320x320` is the best choice for the vast majority of setups. Frigate is specifically designed to compensate for the smaller model by cropping a region of motion from the full frame and zooming into it before running detection, so a `320x320` model is actually _better_ at small and distant objects, not worse. A `640x640` model is slower and uses more resources, and its main benefit is fitting more objects into a single inference when many objects are spread across a large area. Recent versions of Frigate have improved support for `640x640` models, but `320x320` remains the recommended starting point for nearly all setups.

-**Variant size (tiny/small/medium):** Larger variants are gradually more accurate but slower. Whether the difference is noticeable depends on your specific cameras and scenes. A good rule of thumb is to use the largest model your hardware can run without skipping detections, which you can monitor on the <NavPath path="System > Metrics > Cameras" /> page in the UI. Better accuracy only helps if your detector keeps up with the detection load across all cameras.
+**Variant size (tiny/small/medium):** Larger variants are gradually more accurate but slower. Whether the difference is noticeable depends on your specific cameras and scenes. A good rule of thumb is to use the largest model your hardware can run without skipping detections, which you can monitor on the <NavPath path="Health and Metrics > Cameras" /> page in the UI. Better accuracy only helps if your detector keeps up with the detection load across all cameras.

 **Acceptable inference time depends on your hardware.** Inference time alone does not tell the whole story, because different hardware has different capacity. A GPU can run multiple instances of the same model concurrently, so an inference time around 30ms can still keep up with several cameras. A Google Coral runs only a single instance of the model, so it needs a much lower inference time (around 10ms) to keep up.

@@ -93,11 +147,11 @@ The best detection accuracy comes from a model trained on images that look like

 # Officially Supported Detectors

-Frigate provides a number of builtin detector types. By default, Frigate will use a single CPU detector. Other detectors may require additional configuration as described below. When using multiple detectors they will run in dedicated processes, but pull from a common queue of detection requests from across all cameras.
+Frigate provides a number of builtin detector types. By default, Frigate will use a single CPU detector. Other detectors may require additional configuration as described below. Each of a model's devices runs in a dedicated process, and they pull from a common queue of detection requests from the cameras assigned to that model.

 ## Edge TPU Detector

-The Edge TPU detector type runs TensorFlow Lite models utilizing the Google Coral delegate for hardware acceleration. To configure an Edge TPU detector, set the `"type"` attribute to `"edgetpu"`.
+The Edge TPU detector type runs TensorFlow Lite models utilizing the Google Coral delegate for hardware acceleration. To use it, prefix a model's device with `edgetpu`.

 The Edge TPU device can be specified using the `"device"` attribute according to the [Documentation for the TensorFlow Lite Python API](https://coral.ai/docs/edgetpu/multiple-edgetpu/#using-the-tensorflow-lite-python-api). If not set, the delegate will use the first device it finds.

@@ -112,16 +166,15 @@ See [common Edge TPU troubleshooting steps](/troubleshooting/edgetpu) if the Edg
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add**, then set device to `usb`.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral:
-    type: edgetpu
-    device: usb
+models:
+  - devices:
+      - edgetpu:usb
 ```

 </TabItem>
@@ -132,19 +185,16 @@ detectors:
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add** to add multiple detectors, specifying `usb:0` and `usb:1` as the device for each.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown and check each Coral the model should run on.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral1:
-    type: edgetpu
-    device: usb:0
-  coral2:
-    type: edgetpu
-    device: usb:1
+models:
+  - devices:
+      - edgetpu:usb:0
+      - edgetpu:usb:1
 ```

 </TabItem>
@@ -157,16 +207,15 @@ _warning: may have [compatibility issues](https://github.com/blakeblackshear/fri
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add**, then leave the device field empty.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select the **Coral EdgeTPU** entry from the **Hardware** dropdown.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral:
-    type: edgetpu
-    device: ""
+models:
+  - devices:
+      - 'edgetpu:'
 ```

 </TabItem>
@@ -177,16 +226,15 @@ detectors:
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add**, then set device to `pci`.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (PCIe)** from the **Hardware** dropdown.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral:
-    type: edgetpu
-    device: pci
+models:
+  - devices:
+      - edgetpu:pci
 ```

 </TabItem>
@@ -197,19 +245,16 @@ detectors:
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add** to add multiple detectors, specifying `pci:0` and `pci:1` as the device for each.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (PCIe)** from the **Hardware** dropdown and check each Coral the model should run on.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral1:
-    type: edgetpu
-    device: pci:0
-  coral2:
-    type: edgetpu
-    device: pci:1
+models:
+  - devices:
+      - edgetpu:pci:0
+      - edgetpu:pci:1
 ```

 </TabItem>
@@ -220,19 +265,16 @@ detectors:
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and select **EdgeTPU** from the detector type dropdown and click **Add** to add multiple detectors with different device types (e.g., `usb` and `pci`).
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown. USB and PCIe Corals are listed as separate hardware, so mixing the two on one model has to be done in YAML.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors:
-  coral_usb:
-    type: edgetpu
-    device: usb
-  coral_pci:
-    type: edgetpu
-    device: pci
+models:
+  - devices:
+      - edgetpu:usb
+      - edgetpu:pci
 ```

 </TabItem>
@@ -244,9 +286,9 @@ detectors:

 ---

-## Hailo-8
+## Hailo

-This detector is available for use with both Hailo-8 and Hailo-8L AI Acceleration Modules. The integration automatically detects your hardware architecture via the Hailo CLI and selects the appropriate default model if no custom model is specified.
+This detector is available for use with the Hailo-8, Hailo-8L and Hailo-8R AI Acceleration Modules. The integration identifies which of them is attached and selects the matching default model if no custom model is specified.

 See the [installation docs](../frigate/installation.md#hailo-8) for information on configuring the Hailo hardware.

@@ -256,16 +298,22 @@ If no custom model is provided, the Hailo detector downloads a default model fro

 :::

+:::info
+
+The HailoRT runtime is not part of the Frigate image. It is downloaded and installed into `/config/.local` the first time a Hailo detector is configured, verified against pinned checksums, and updated automatically when a Frigate release pins a new version. If the container has no internet access, see [Detector runtimes](/frigate/network_requirements#detector-runtimes) for how to provide the files yourself.
+
+:::
+
 ### Configuration {#configuration-hailo}

 When configuring the Hailo detector, you have two options to specify the model: a local **path** or a **URL**.
 If both are provided, the detector will first check for the model at the given local path. If the file is not found, it will download the model from the specified URL. The model file is cached under `/config/model_cache/hailo`.

-<ModelConfigDropdown detectorTitle="Hailo-8/Hailo-8L" models={objectDetectorsModels.hailo8l.models} />
+<ModelConfigDropdown detectorTitle="Hailo" models={objectDetectorsModels.hailo.models} />

 For additional ready-to-use models, please visit: https://github.com/hailo-ai/hailo_model_zoo

-Hailo8 supports all models in the Hailo Model Zoo that include HailoRT post-processing. You're welcome to choose any of these pre-configured models for your implementation.
+Hailo supports all models in the Hailo Model Zoo that include HailoRT post-processing. You're welcome to choose any of these pre-configured models for your implementation.

 > **Note:**
 > The config.path parameter can accept either a local file path or a URL ending with .hef. When provided, the detector will first check if the path is a local file path. If the file exists locally, it will use it directly. If the file is not found locally or if a URL was provided, it will attempt to download the model from the specified URL.
@@ -274,7 +322,7 @@ Hailo8 supports all models in the Hailo Model Zoo that include HailoRT post-proc

 ## OpenVINO Detector

-The OpenVINO detector type runs an OpenVINO IR model on AMD and Intel CPUs, Intel GPUs and Intel NPUs. To configure an OpenVINO detector, set the `"type"` attribute to `"openvino"`.
+The OpenVINO detector type runs an OpenVINO IR model on AMD and Intel CPUs, Intel GPUs and Intel NPUs. To use it, prefix a model's device with `openvino`.

 The OpenVINO device to be used is specified using the `"device"` attribute according to the naming conventions in the [Device Documentation](https://docs.openvino.ai/2025/openvino-workflow/running-inference/inference-devices-and-modes.html). The most common devices are `CPU`, `GPU`, or `NPU`.

@@ -287,17 +335,24 @@ OpenVINO is supported on 6th Gen Intel platforms (Skylake) and newer. It will al
 When using many cameras one detector may not be enough to keep up. Multiple detectors can be defined assuming GPU resources are available. An example configuration would be:

 ```yaml
-detectors:
-  ov_0:
-    type: openvino
-    device: GPU # or NPU
-  ov_1:
-    type: openvino
-    device: GPU # or NPU
+models:
+  - devices:
+      - openvino:GPU # or NPU
+      - openvino:GPU # or NPU
 ```

 :::

+### Intel NPU host requirements {#intel-npu-requirements}
+
+The NPU device must be passed into the container by adding `/dev/accel:/dev/accel` to the `devices` section of your compose file. Frigate grants the runtime user access to the device automatically; see [hardware device access](/configuration/non_root#hardware-device-access) if you manage device permissions yourself.
+
+The NPU firmware is loaded by the host kernel and is not part of the Frigate image. Everything else the NPU needs is bundled in the container, so host NPU libraries should never be mounted in.
+
+Frigate bundles a specific version of Intel's [linux-npu-driver](https://github.com/intel/linux-npu-driver/releases), and the host firmware must come from that release or a newer one. Firmware older than the bundled driver may fail with `MAPPED_INFERENCE_VERSION is NOT compatible with the ELF`, where `Expected` is the version the firmware supports and `received` is the version the bundled compiler produced. Distributions often package older firmware than the driver Frigate ships, so check the build date on the host with `sudo dmesg | grep -i vpu` and update it there if needed.
+
+Intel NPUs cannot be used under Home Assistant OS, which does not include the NPU firmware.
+
 ### Configuration {#configuration-openvino}

 <ModelConfigDropdown detectorTitle="OpenVINO" models={objectDetectorsModels.openvino.models} />
@@ -306,6 +361,12 @@ detectors:

 ## Apple Silicon detector

+:::warning
+
+The Apple Silicon detector client is being reworked. Its extra options no longer have a place in the config, so only the endpoint carried in the device string is honored right now, and `request_timeout_ms` and `linger_ms` are ignored. Anything else is dropped when your config is migrated.
+
+:::
+
 The NPU in Apple Silicon can't be accessed from within a container, so the [Apple Silicon detector client](https://github.com/frigate-nvr/apple-silicon-detector) must first be setup. It is recommended to use the Frigate docker image with `-standard-arm64` suffix, for example `ghcr.io/blakeblackshear/frigate:stable-standard-arm64`.

 ### Setup {#setup-apple-silicon}
@@ -446,11 +507,10 @@ If the correct build is used for your GPU then the GPU will be detected and used
 When using many cameras one detector may not be enough to keep up. Multiple detectors can be defined assuming GPU resources are available. An example configuration would be:

 ```yaml
-detectors:
-  onnx_0:
-    type: onnx
-  onnx_1:
-    type: onnx
+models:
+  - devices:
+      - onnx
+      - onnx
 ```

 :::
@@ -463,7 +523,7 @@ detectors:

 ## CPU Detector (not recommended)

-The CPU detector type runs a TensorFlow Lite model utilizing the CPU without hardware acceleration. It is recommended to use a hardware accelerated detector type instead for better performance. To configure a CPU based detector, set the `"type"` attribute to `"cpu"`.
+The CPU detector type runs a TensorFlow Lite model utilizing the CPU without hardware acceleration. It is recommended to use a hardware accelerated detector type instead for better performance. To use it, set a model's device to `cpu`.

 :::danger

@@ -473,7 +533,7 @@ The CPU detector is not recommended for general use. If you do not have GPU or E

 The number of threads used by the interpreter can be specified using the `"num_threads"` attribute, and defaults to `3.`

-A TensorFlow Lite model is provided in the container at `/cpu_model.tflite` and is used by this detector type by default. To provide your own model, bind mount the file into the container and provide the path with `model.path`.
+A TensorFlow Lite model is provided in the container at `/cpu_model.tflite` and is used by this detector type by default. To provide your own model, bind mount the file into the container and provide the path with the model's `path`.

 ### Configuration {#configuration-cpu}

@@ -481,24 +541,6 @@ A TensorFlow Lite model is provided in the container at `/cpu_model.tflite` and

 When using CPU detectors, you can add one CPU detector per camera. Adding more detectors than the number of cameras should not improve performance.

-## Deepstack / CodeProject.AI Server Detector
-
-The Deepstack / CodeProject.AI Server detector for Frigate allows you to integrate Deepstack and CodeProject.AI object detection capabilities into Frigate. CodeProject.AI and DeepStack are open-source AI platforms that can be run on various devices such as the Raspberry Pi, Nvidia Jetson, and other compatible hardware. It is important to note that the integration is performed over the network, so the inference times may not be as fast as native Frigate detectors, but it still provides an efficient and reliable solution for object detection and tracking.
-
-### Setup {#setup-deepstack}
-
-To get started with CodeProject.AI, visit their [official website](https://www.codeproject.com/Articles/5322557/CodeProject-AI-Server-AI-the-easy-way) to follow the instructions to download and install the AI server on your preferred device. Detailed setup instructions for CodeProject.AI are outside the scope of the Frigate documentation.
-
-To integrate CodeProject.AI into Frigate, configure the detector as follows:
-
-### Configuration {#configuration-deepstack}
-
-<ModelConfigDropdown detectorTitle="DeepStack" models={objectDetectorsModels.deepstack.models} />
-
-Replace `<your_codeproject_ai_server_ip>` and `<port>` with the IP address and port of your CodeProject.AI server.
-
-To verify that the integration is working correctly, start Frigate and observe the logs for any error messages related to CodeProject.AI. Additionally, you can check the Frigate web interface to see if the objects detected by CodeProject.AI are being displayed and tracked properly.
-
 # Community Supported Detectors

 ## MemryX MX3
@@ -509,6 +551,12 @@ See the [installation docs](../frigate/installation.md#memryx-mx3) for informati

 To configure a MemryX detector, simply set the `type` attribute to `memryx` and follow the configuration guide below.

+:::info
+
+The MemryX SDK is not part of the Frigate image. It is downloaded and installed into `/config/.local` the first time a MemryX detector is configured, verified against pinned checksums, and updated automatically when a Frigate release pins a new version. If the container has no internet access, see [Detector runtimes](/frigate/network_requirements#detector-runtimes) for how to provide the files yourself.
+
+:::
+
 ### Configuration {#configuration-memryx}

 <ModelConfigDropdown detectorTitle="MemryX" models={objectDetectorsModels.memryx.models} />
@@ -545,7 +593,7 @@ For detailed instructions on compiling models, refer to the [MemryX Compiler](ht

 3. Depending on the model, the compiler may also generate a cropped post-processing network. If present, it will be named with the suffix `_post.onnx`.

-4. Bind-mount the `.zip` file into the container and specify its path using `model.path` in your config.
+4. Bind-mount the `.zip` file into the container and specify its path using the model's `path` in your config.

 5. Update `labelmap_path` to match your custom model's labels.

@@ -565,6 +613,63 @@ For detailed instructions on compiling models, refer to the [MemryX Compiler](ht

 ---

+## DEEPX NPU
+
+This detector is available for use with the DEEPX NPU, both the DX-M1 M.2 module and the DX-M1M on the Sixfab AI HAT+ for the Raspberry Pi 5. The configuration below applies unchanged to either form factor. DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
+
+See the [installation docs](../frigate/installation.md#deepx-npu) for information on installing the DEEPX kernel driver and runtime on the host and passing the NPU through to the container.
+
+To run a model on a DEEPX NPU, list a `deepx` device on that model.
+
+:::info
+
+The DX-RT Python bindings are not part of the Frigate image. They are downloaded and installed into `/config/.local` the first time a DEEPX device is configured, verified against pinned checksums, and updated automatically when a Frigate release pins a new version. If the container has no internet access, see [Detector runtimes](/frigate/network_requirements#detector-runtimes) for how to provide the files yourself.
+
+:::
+
+### Configuration {#configuration-deepx}
+
+<ModelConfigDropdown detectorTitle="DEEPX" models={objectDetectorsModels.deepx.models} />
+
+Frigate does not bundle a model for this detector. Models must be compiled to DEEPX's `.dxnn` format. Two model types are supported:
+
+- `yolo-generic` for YOLO object detection models, the recommended default. The detector reads the model's output layout from the compiled file, so anchor-based, anchor-free and NMS-in-head models all work with the same configuration, as do models compiled with DEEPX's Post-Processing Unit (PPU) support.
+- `yolox` for YOLOX models compiled without PPU support, whose raw head needs Frigate's YOLOX decoder. A YOLOX model compiled with PPU support works under either `yolox` or `yolo-generic`.
+
+The quickest way to get one is the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo), which publishes pre-compiled `.dxnn` files for a range of YOLO object detection models. Download the `.dxnn`, bind-mount it into the container, and point the model's `path` at it. Alternatively, compile your own model with the DX-COM compiler. The recommended starting point is `yolox-s_640x640_ppu.dxnn`, the fastest ModelZoo model measured through Frigate:
+
+```yaml
+models:
+  - devices:
+      - deepx:PCIe:0
+    path: /config/model_cache/deepx/yolox-s_640x640_ppu.dxnn
+    labelmap_path: /labelmap/coco-80.txt
+    model_type: yolo-generic
+    width: 640
+    height: 640
+```
+
+For PPU models, use a `.dxnn` compiled with DX-COM 2.4.0 or later. Frigate reads the PPU head layout the compiler writes into the file and refuses to load a PPU model without it.
+
+`model_type` must be set to `yolo-generic` or `yolox` to match the model; `yolo-generic` is the recommended default unless the model is a raw YOLOX export. Frigate defaults it to `ssd`, which this detector does not support, so the detector refuses to start on a model that leaves it unset.
+
+`width` and `height` must match the resolution the model was compiled for. Quantization parameters are baked into the `.dxnn` file at compile time, so no normalization is applied on the host and Frigate's default `input_tensor`, `input_pixel_format`, and `input_dtype` values do not need to be overridden.
+
+A DEEPX device is `PCIe:<index>`, as reported on the detector settings page. The NPU daemon multiplexes across processes, so the same device may be listed more than once to run additional inference processes against it:
+
+```yaml
+models:
+  - devices:
+      - deepx:PCIe:0
+      - deepx:PCIe:0
+```
+
+#### Label maps
+
+The object detection models in the DEEPX ModelZoo are trained on the standard 80-class COCO label set, so `labelmap_path` must be set to `/labelmap/coco-80.txt`. Frigate's default label map uses an extended 91-class COCO scheme, and leaving it in place will cause detections to be reported as the wrong object type. For `yolo-generic` models the label map is also what the detector uses to tell the output layout, so a label map with the wrong number of classes is reported as an error at startup.
+
+---
+
 ## NVidia TensorRT Detector

 Nvidia Jetson devices may be used for object detection using the TensorRT libraries. Due to the size of the additional libraries, this detector is only provided in images with the `-tensorrt-jp6` tag suffix, e.g. `ghcr.io/blakeblackshear/frigate:stable-tensorrt-jp6`. This detector is designed to work with Yolo models for object detection.
@@ -675,13 +780,10 @@ If no custom model is provided, the RKNN detector downloads a default model from
 When using many cameras one detector may not be enough to keep up. Multiple detectors can be defined assuming NPU resources are available. An example configuration would be:

 ```yaml
-detectors:
-  rknn_0:
-    type: rknn
-    num_cores: 0
-  rknn_1:
-    type: rknn
-    num_cores: 0
+models:
+  - devices:
+      - rknn:0
+      - rknn:0
 ```

 :::
@@ -755,87 +857,6 @@ Explanation of the parameters:
  - **example**: Specifying `output_name = "frigate-{quant}-{input_basename}-{soc}-v{tk_version}"` could result in a model called `frigate-i8-my_model-rk3588-v2.3.0.rknn`.
 - `config`: Configuration passed to `rknn-toolkit2` for model conversion. For an explanation of all available parameters have a look at section "2.2. Model configuration" of [this manual](https://github.com/MarcA711/rknn-toolkit2/releases/download/v2.3.2/03_Rockchip_RKNPU_API_Reference_RKNN_Toolkit2_V2.3.2_EN.pdf).

-## DeGirum
-
-DeGirum is a detector that can use any type of hardware listed on [their website](https://hub.degirum.com). DeGirum can be used with local hardware through a DeGirum AI Server, or through the use of `@local`. You can also connect directly to DeGirum's AI Hub to run inferences. **Please Note:** This detector _cannot_ be used for commercial purposes.
-
-### Configuration {#configuration-degirum}
-
-#### AI Server Inference
-
-Before starting with the config file for this section, you must first launch an AI server. DeGirum has an AI server ready to use as a docker container. Add this to your `docker-compose.yml` to get started:
-
-```yaml
-degirum_detector:
-  container_name: degirum
-  image: degirum/aiserver:latest
-  privileged: true
-  ports:
-    - "8778:8778"
-```
-
-All supported hardware will automatically be found on your AI server host as long as relevant runtimes and drivers are properly installed on your machine. Refer to [DeGirum's docs site](https://docs.degirum.com/pysdk/runtimes-and-drivers) if you have any trouble.
-
-Once completed, configure the detector as follows:
-
-<ModelConfigDropdown detectorTitle="DeGirum" models={objectDetectorsModels.degirumAiServer.models} />
-
-Setting up a model in the `config.yml` is similar to setting up an AI server.
-You can set it to:
-
- A model listed on the [AI Hub](https://hub.degirum.com), given that the correct zoo name is listed in your detector
-  - If this is what you choose to do, the correct model will be downloaded onto your machine before running.
- A local directory acting as a zoo. See DeGirum's docs site [for more information](https://docs.degirum.com/pysdk/user-guide-pysdk/organizing-models#model-zoo-directory-structure).
- A path to some model.json.
-
-```yaml
-model:
-  path: ./mobilenet_v2_ssd_coco--300x300_quant_n2x_orca1_1 # directory to model .json and file
-  width: 300 # width is in the model name as the first number in the "int"x"int" section
-  height: 300 # height is in the model name as the second number in the "int"x"int" section
-  input_pixel_format: rgb/bgr # look at the model.json to figure out which to put here
-```
-
-#### Local Inference
-
-It is also possible to eliminate the need for an AI server and run the hardware directly. The benefit of this approach is that you eliminate any bottlenecks that occur when transferring prediction results from the AI server docker container to the frigate one. However, the method of implementing local inference is different for every device and hardware combination, so it's usually more trouble than it's worth. A general guideline to achieve this would be:
-
-1. Ensuring that the frigate docker container has the runtime you want to use. So for instance, running `@local` for Hailo means making sure the container you're using has the Hailo runtime installed.
-2. To double check the runtime is detected by the DeGirum detector, make sure the `degirum sys-info` command properly shows whatever runtimes you mean to install.
-3. Create a DeGirum detector in your configuration.
-
-<ModelConfigDropdown detectorTitle="DeGirum" models={objectDetectorsModels.degirumLocal.models} />
-
-Once `degirum_detector` is setup, you can choose a model through 'model' section in the `config.yml` file.
-
-```yaml
-model:
-  path: mobilenet_v2_ssd_coco--300x300_quant_n2x_orca1_1
-  width: 300 # width is in the model name as the first number in the "int"x"int" section
-  height: 300 # height is in the model name as the second number in the "int"x"int" section
-  input_pixel_format: rgb/bgr # look at the model.json to figure out which to put here
-```
-
-#### AI Hub Cloud Inference
-
-If you do not possess whatever hardware you want to run, there's also the option to run cloud inferences. Do note that your detection fps might need to be lowered as network latency does significantly slow down this method of detection. For use with Frigate, we highly recommend using a local AI server as described above. To set up cloud inferences,
-
-1. Sign up at [DeGirum's AI Hub](https://hub.degirum.com).
-2. Get an access token.
-3. Create a DeGirum detector in your configuration.
-
-<ModelConfigDropdown detectorTitle="DeGirum" models={objectDetectorsModels.degirumCloud.models} />
-
-Once `degirum_detector` is setup, you can choose a model through 'model' section in the `config.yml` file.
-
-```yaml
-model:
-  path: mobilenet_v2_ssd_coco--300x300_quant_n2x_orca1_1
-  width: 300 # width is in the model name as the first number in the "int"x"int" section
-  height: 300 # height is in the model name as the second number in the "int"x"int" section
-  input_pixel_format: rgb/bgr # look at the model.json to figure out which to put here
-```
-
 ## AXERA

 Hardware accelerated object detection is supported on the following SoCs:
@@ -853,6 +874,12 @@ The AXEngine detector downloads its default model from HuggingFace on first star

 :::

+:::info
+
+The AXEngine python package is not part of the Frigate image. It is downloaded and installed into `/config/.local` the first time an AXEngine detector is configured, verified against a pinned checksum, and updated automatically when a Frigate release pins a new version. If the container has no internet access, see [Detector runtimes](/frigate/network_requirements#detector-runtimes) for how to provide the file yourself.
+
+:::
+
 ### Configuration {#configuration-axengine}

 When configuring the AXEngine detector, you have to specify the model name.
@@ -126,7 +126,7 @@ Only the fields you explicitly set in a profile override are applied. All other

 ## Activating Profiles

-Profiles can be activated and deactivated via the Frigate UI, [MQTT](/integrations/mqtt#frigateprofileset), or the Home Assistant integration.
+Profiles can be activated and deactivated via the Frigate UI, [MQTT](/integrations/mqtt#frigateprofileset), the [HTTP API](../integrations/api/camera-set-camera-camera-name-set-feature-sub-command-put.api.mdx), or the Home Assistant integration.

 In the Frigate UI, open the Settings cog and select **Profiles** from the submenu to see all defined profiles. From there you can activate any profile or deactivate the current one. The active profile is indicated in the UI so you always know which profile is in effect.

@@ -9,7 +9,7 @@ import NavPath from "@site/src/components/NavPath";

 Recordings can be enabled and are stored at `/media/frigate/recordings`. The folder structure for the recordings is `YYYY-MM-DD/HH/<camera_name>/MM.SS.mp4` in **UTC time**. These recordings are written directly from your camera stream without re-encoding. Each camera supports a configurable retention policy. Frigate chooses the largest matching retention value between the recording retention and the tracked object retention when determining if a recording should be removed.

-New recording segments are written from the camera stream to cache, they are only moved to disk if they match the setup recording retention policy.
+New recording segments are written from the camera stream to cache, they are only moved to disk if they pass a validation check and match the setup recording retention policy.

 :::tip

@@ -275,6 +275,165 @@ record:

 This configuration will retain recording segments that overlap with alerts and detections for 10 days. Because multiple tracked objects can reference the same recording segments, this avoids storing duplicate footage for overlapping tracked objects and reduces overall storage needs.

+## Sub Stream Recording
+
+In addition to the main recording stream, Frigate can record a second, lower quality stream for each camera. This serves two purposes:
+
+- **Quality selection during playback**: A quality selector (`Auto`, `Original`, or `Low`) appears in History view for cameras with sub stream recording enabled. `Original` and `Low` play only that stream's recordings. Time ranges where the selected stream has no footage are skipped during playback, and the selector notes when the selected stream has no recordings at all in the viewed time range. With `Auto` (the default), playback prefers the original quality and automatically falls back to the low quality stream when the connection cannot keep up, or for time ranges where the original recordings have expired. The selector shows each stream's video codec and audio details beneath the options; footage recorded by older Frigate versions shows no details.
+- **Quality selection when exporting**: A `Quality` selector (`Auto`, `Original`, or `Low`) is available for cameras with sub stream recording enabled. See [exporting](#exporting-a-camera-that-records-two-streams) for details on each option.
+- **Extended retention**: Sub stream recordings have their own retention settings, fully independent of the main recordings. By giving the low quality recordings a longer retention period, you can keep weeks or months of low quality history using a fraction of the storage, and that history remains playable after the main recordings expire. Playback falls back to the low quality recordings automatically, and the timeline shows a muted treatment for time ranges where only low quality footage remains. Timeline previews are kept for as long as either stream still has recordings, so scrubbing works across the whole retained history.
+
+### Configuring sub stream recording
+
+Sub stream recording uses the `record_sub` input role. This role can be assigned to the same input as `detect`, so in the common case where detect already uses the camera's sub stream, no additional camera connection is needed. Like the main recording stream, sub stream segments are copied directly from the camera stream without re-encoding, so the recording quality is determined by the source stream.
+
+The following examples keep 7 days of full quality continuous recordings and 60 days of low quality continuous recordings:
+
+<ConfigTabs>
+<TabItem value="ui">
+
+Navigate to <NavPath path="Settings > Camera configuration > Streams (FFmpeg)" /> and select the camera.
+
+- In **Camera inputs**, enable the **Record (Sub Stream)** role on the stream you want to record at low quality, commonly the same stream that has the **Detect** role. Only one stream may have this role, and it cannot be assigned to the same stream as the **Record** role.
+
+Navigate to <NavPath path="Settings > Camera configuration > Recording" /> and select the camera.
+
+- Set **Enable recording** to on
+- Set **Continuous retention > Retention days** to `7`
+- Set **Sub stream recording > Enable sub stream recording** to on
+- Set **Sub stream recording > Sub stream continuous retention > Retention days** to `60`
+
+The camera setup wizard also offers the **Record (Sub Stream)** role when assigning stream roles for a newly added camera.
+
+</TabItem>
+<TabItem value="yaml">
+
+```yaml
+cameras:
+  front_door:
+    ffmpeg:
+      inputs:
+        - path: rtsp://camera/main
+          roles:
+            - record
+        - path: rtsp://camera/sub
+          roles:
+            - detect
+            - record_sub
+    record:
+      enabled: true
+      continuous:
+        days: 7
+      sub:
+        enabled: true
+        continuous:
+          days: 60
+```
+
+If your camera does not provide a suitable sub stream (or the sub stream is already used at a resolution you don't want to record), you can use a go2rtc transcode as the source for `record_sub` instead:
+
+```yaml
+go2rtc:
+  streams:
+    front_door: rtsp://camera/main
+    front_door_lq: ffmpeg:front_door#video=h264#width=854#hardware
+
+cameras:
+  front_door:
+    ffmpeg:
+      inputs:
+        - path: rtsp://127.0.0.1:8554/front_door
+          input_args: preset-rtsp-restream
+          roles:
+            - detect
+            - record
+        - path: rtsp://127.0.0.1:8554/front_door_lq
+          input_args: preset-rtsp-restream
+          roles:
+            - record_sub
+    record:
+      enabled: true
+      continuous:
+        days: 7
+      sub:
+        enabled: true
+        continuous:
+          days: 60
+```
+
+</TabItem>
+</ConfigTabs>
+
+The `record.sub` config supports the same retention structure as the main recording config: `continuous`, `motion`, `alerts`, and `detections` each with their own `days` (and `mode` for alerts and detections). The pre-capture and post-capture windows for alerts and detections are taken from the main `record.alerts` and `record.detections` config. Extending `sub.alerts.days` or `sub.detections.days` beyond the main values also keeps those review items visible in the review timeline for the longer window, with playback falling back to the low quality stream once the main recordings expire.
+
+:::note
+
+Recording must be enabled (`record.enabled`) for sub stream recording to run, and Frigate will fail to start if `record.sub.enabled` is set without a `record_sub` role assigned to one of the camera's inputs.
+
+:::
+
+### How Auto picks a quality
+
+`Auto` measures throughput on every segment download and compares it against the original stream's bitrate (computed from the recorded footage itself). Playback drops to the low quality stream when any of these happen:
+
+- A freeze lasts 4 seconds (10 seconds when it starts within 2 seconds of a seek, since the seek target is rarely buffered), or freezes total 7 seconds within the last minute.
+- 3 downloads in a row measure below the original bitrate plus 10%, dropping quality before a stall ever becomes visible.
+- No first frame appears within 10 seconds, or loading fails outright.
+
+Playback returns to full quality only when measured throughput exceeds the original bitrate by 50%, checked continuously while playing the low quality stream and again at each new hour. The asymmetric thresholds (1.1x to drop, 1.5x to return) keep a borderline connection from switching back and forth.
+
+The most recent measurement is remembered on the device: a connection last measured below the original bitrate (or below 3 Mbps when the bitrate is not yet known) starts playback on the low quality stream so a first frame appears immediately, then upgrades within a few segments if the speed allows.
+
+The quality selector shows which stream Auto is currently playing and why. A browser with Data Saver enabled stays on the low quality stream, a browser that cannot decode the original stream's codec (for example H.265 without HEVC support) plays the low quality stream for that camera, and pinning `Original` or `Low` bypasses Auto entirely.
+
+### Sub stream output args
+
+By default the sub stream is recorded with the same [output args](/configuration/ffmpeg_presets#output-args-presets) as the main recording stream, so it inherits any customization made to `ffmpeg.output_args.record`. Setting `ffmpeg.output_args.record_sub` gives the sub stream its own args instead. Like all `ffmpeg` config, this can be set globally or per camera.
+
+The most common reason to set this is a pair of streams whose audio differs. Many cameras send AAC on the main stream but PCM on the sub stream, and PCM cannot be copied into an mp4 recording. Copying the main stream's audio avoids re-encoding audio that is already AAC, while the sub stream still needs to be transcoded:
+
+```yaml
+ffmpeg:
+  output_args:
+    # main stream audio is already AAC, so copy it
+    record: preset-record-generic-audio-copy
+    # sub stream audio is PCM, so transcode it to AAC
+    record_sub: preset-record-generic-audio-aac
+```
+
+Other reasons to set this are recording a sub stream whose codec needs a different preset than the main stream, such as `preset-record-mjpeg`, or forcing a matching audio sample rate across the two streams with manual args ending in `-c:a aac -ar 16000`.
+
+:::warning
+
+Avoid removing audio from only one of the two streams (for example with `-an`). When one stream has audio and the other does not, playback of time ranges that combine both qualities is silent, so stripping audio from the sub stream also silences the merged timeline.
+
+:::
+
+### Which stream do features use?
+
+As a general rule, features that read recordings prefer the main stream and fall back to the sub stream for time ranges where the main recordings have expired. Analytics features use only the main stream.
+
+| Feature                                                                                  | Stream used                                                                                                   |
+| ---------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- |
+| Recording playback (History and Review)                                                  | Both (main preferred with sub fallback by default), or exactly one stream when a quality is selected manually |
+| Tracking details and Explore clip playback                                               | Main, falling back to sub where the main recordings have expired                                              |
+| Exports                                                                                  | Both (main preferred with sub fallback by default), or exactly one stream when a quality is selected in the export dialog |
+| Clip downloads                                                                           | Main; sub is used when no main recordings remain in the range (streams are never mixed in one file)           |
+| Frames grabbed from a recording in History (download snapshot, submit frame to Frigate+) | Main preferred, sub fallback                                                                                  |
+| Audio extraction (e.g., transcription)                                                   | Main preferred, sub fallback                                                                                  |
+| Motion search                                                                            | Main only                                                                                                     |
+| Review timeline motion data                                                              | Main only                                                                                                     |
+| Storage usage statistics                                                                 | Both streams counted, and listed separately per camera                                                        |
+
+This table covers only features that read recordings from disk. Tracked object snapshots and thumbnails (the images shown in Explore and sent with notifications, and the images submitted to Frigate+ from a tracked object) are captured live from the `detect` stream as the object is tracked, never from recordings, so sub stream recording does not affect them.
+
+### Trade-offs
+
+- Recording a second stream increases overall storage use. The increase is typically small relative to the main recordings, since the low quality stream is much smaller. Both streams are cached before being written to disk, so cache use goes up as well. See [the `/tmp/cache` area is separate](#the-tmpcache-area-is-separate) if you start seeing `No space left on device` errors after enabling it.
+- The go2rtc transcode approach continuously encodes the low quality stream, which uses CPU or GPU resources. This cost only applies to the transcode path; recording the camera's native sub stream does not re-encode. See the [go2rtc hardware acceleration documentation](https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#source-ffmpeg) for accelerating the transcode.
+- Many camera sub streams do not include audio. If the source stream has no audio, the low quality recordings will not have audio.
+- **Matching video codecs and audio settings between the two streams gives the smoothest playback.** When playback combines both qualities on one timeline (the default `Auto` behavior: for example original quality during events with low quality in between, or low quality history after the original recordings expire) and the streams use different video codecs or audio settings, for example H.265 on the main stream and H.264 on the sub stream, or 16 kHz audio on one and 8 kHz on the other, playback still works: Frigate inserts a decoder reset at each quality transition, which can cause a barely-perceptible pause there. Configuring both streams in the camera's firmware to use the same video codec, audio codec, and sample rate makes transitions fully seamless, and a mismatched audio sample rate can also be corrected with [sub stream output args](#sub-stream-output-args). If one stream has audio and the other does not, combined time ranges play **without audio**; selecting a single quality with the playback selector always keeps that stream's audio.
+
 ## Can I have "continuous" recordings, but only at certain times?

 Using Frigate UI, Home Assistant, or MQTT, cameras can be automated to only record in certain situations or at certain times.
@@ -291,7 +450,7 @@ For advanced use cases, the [custom export HTTP API](../integrations/api/export-
 POST /export/custom/{camera_name}/start/{start_time}/end/{end_time}
 ```

-The request body accepts `ffmpeg_input_args` and `ffmpeg_output_args` to control encoding, frame rate, filters, and other FFmpeg options. If neither is provided, Frigate defaults to time-lapse output settings (25x speed, 30 FPS).
+The request body accepts `ffmpeg_input_args` and `ffmpeg_output_args` to control encoding, frame rate, filters, and other FFmpeg options. If neither is provided, Frigate defaults to time-lapse output settings (25x speed, 30 FPS) with audio removed (`-an`). When providing your own `ffmpeg_input_args`, include `-an` if you want audio stripped from the export.

 The following example exports a time-lapse at 60x speed with 25 FPS:

@@ -342,7 +501,7 @@ Media files (event snapshots, event thumbnails, review thumbnails, previews, exp

 Normal operation may leave small numbers of orphaned files until Frigate's scheduled cleanup, but crashes, configuration changes, or upgrades may cause more orphaned files that Frigate does not clean up. This feature checks the file system for media files and removes any that are not referenced in the database.

-The Maintenance pane in the Frigate UI or an API endpoint `POST /api/media/sync` can be used to trigger a media sync. When using the API, a job ID is returned and the operation continues on the server. Status can be checked with the `/api/media/sync/status/{job_id}` endpoint.
+The Maintenance pane in the Frigate UI or an API endpoint `POST /api/media/sync` can be used to trigger a media sync. When using the API, a job ID is returned and the operation continues on the server. Status can be checked with the `/api/media/sync/status/{job_id}` endpoint. Results include the disk space reclaimed, or with `dry_run: true`, the space that would be reclaimed.

 Setting `verbose: true` writes a detailed report of every orphaned file and database entry to `/config/media_sync/<job_id>.txt`. For recordings, the report separates orphaned database entries (DB records whose files are missing from disk) from orphaned files (files on disk with no corresponding database record).

@@ -358,7 +517,7 @@ The storage usage Frigate reports will not exactly match what the operating syst

 ### How Frigate measures recording usage

-The **Recordings** value on the Storage Metrics page (<NavPath path="System > Storage" />), and the per-camera **Camera Storage** breakdown, is the sum of the recording segment sizes Frigate has written, taken from Frigate's database. It is **not** computed by a scan of the disk. Frigate tracks usage this way by design: repeatedly walking the entire drive to total its size would keep hard drives spun up and add unnecessary I/O.
+The **Recordings** value on the Storage Metrics page (<NavPath path="Health and Metrics > Storage" />), and the per-camera **Camera Storage** breakdown, is the sum of the recording segment sizes Frigate has written, taken from Frigate's database. It is **not** computed by a scan of the disk. Frigate tracks usage this way by design: repeatedly walking the entire drive to total its size would keep hard drives spun up and add unnecessary I/O.

 The disk **total** shown beside it, and the free-space figure Frigate uses to decide when to delete recordings, instead come from the operating system's report for the whole filesystem mounted at `/media/frigate`. As a result, the **Unused** value on the page is _total disk capacity minus Frigate's recordings_, not the drive's real free space, which will be lower whenever anything else is stored on the disk.

@@ -197,7 +197,7 @@ For cameras that support two-way talk, go2rtc will automatically establish an au
 To prevent this, you must configure two separate stream instances:

 1. One stream instance with `#backchannel=0` for Frigate's viewing, recording, and detection (prevents go2rtc from establishing the blocking backchannel)
-2. A second stream instance without `#backchannel=0` for two-way talk functionality (can be used by Frigate's WebRTC viewer or other applications)
+2. A second stream instance with no `#` parameters at all for two-way talk functionality (can be used by Frigate's WebRTC viewer or other applications)

 Configuration example:

@@ -215,13 +215,15 @@ In this configuration:
 - `front_door` stream is used by Frigate for viewing, recording, and detection. The `#backchannel=0` parameter prevents go2rtc from establishing the audio output backchannel, so it won't block two-way talk access.
 - `front_door_twoway` stream is used for two-way talk functionality. This stream can be used by Frigate's WebRTC viewer when two-way talk is enabled, or by other applications (like Home Assistant Advanced Camera Card) that need access to the camera's audio output channel.

+Any `#` parameter on a bare `rtsp://` source disables the backchannel unless the URL explicitly contains `#backchannel=1`. A two-way talk stream with something like `#video=h264` on it silently loses two-way audio, and Frigate will report that two-way talk is unavailable for that stream.
+
 ## Security: Restricted Stream Sources

 For security reasons, the `echo:`, `expr:`, and `exec:` stream sources are disabled by default in go2rtc. These sources allow arbitrary command execution and can pose security risks if misconfigured.

 If you attempt to use these sources in your configuration, the streams will be removed and an error message will be printed in the logs.

-To enable these sources, you must set the environment variable `GO2RTC_ALLOW_ARBITRARY_EXEC=true`. This can be done in your Docker Compose file or container environment:
+To enable these sources, you must set the environment variable `GO2RTC_ALLOW_ARBITRARY_EXEC=true`. This can be done in your Docker Compose file or container environment, or for Home Assistant App users with the `go2rtc_allow_arbitrary_exec` option in the App's configuration. The `environment_vars` section of the Frigate config can't enable it:

 ```yaml
 environment:
@@ -121,6 +121,31 @@ cameras:
 </TabItem>
 </ConfigTabs>

+## Categorizing manual events
+
+Events created with the [create manual event API](../integrations/api/create-event-events-camera-name-label-create-post.api.mdx) are categorized with the same label lists, using the label from the request path:
+
+1. If alerts are enabled and the label is listed in `review -> alerts -> labels`, the review item is an alert.
+2. Otherwise, if detections are enabled and the label is listed in `review -> detections -> labels`, the review item is a detection.
+3. If the label is in neither list, the review item is an alert, or no review item is created if alerts are disabled.
+
+This means manual events are alerts unless you explicitly list their label as a detection label. For example, to have PIR sensors create detections instead of alerts, post to `/api/events/front_door/pir_sensor/create` with the following config:
+
+```yaml {5-7}
+cameras:
+  front_door:
+    review:
+      detections:
+        labels:
+          - pir_sensor
+```
+
+:::note
+
+Required zones do not apply to manual events, since they are created through the API rather than by the object tracker. Setting `review -> alerts -> labels` to an empty list also does not stop manual events from becoming alerts, as a label in neither list still falls back to an alert.
+
+:::
+
 ## Restricting review items to specific zones

 By default a review item will be created if any `review -> alerts -> labels` and `review -> detections -> labels` are detected anywhere in the camera frame. You will likely want to configure review items to only be created when the object enters an area of interest, [see the zone docs for more information](./zones.md#restricting-alerts-and-detections-to-specific-zones)
@@ -9,7 +9,7 @@ import NavPath from "@site/src/components/NavPath";

 # TLS

-Frigate's integrated NGINX server supports TLS certificates. By default Frigate will generate a self signed certificate that will be used for port 8971. Frigate is designed to make it easy to use whatever tool you prefer to manage certificates.
+Frigate's integrated NGINX server supports TLS certificates. By default Frigate will generate a self signed certificate that will be used for port 8971, stored in `/config/tls` so it survives container recreation. Frigate is designed to make it easy to use whatever tool you prefer to manage certificates.

 Frigate is often running behind a reverse proxy that manages TLS certificates for multiple services. You will likely need to set your reverse proxy to allow self signed certificates or you can disable TLS in Frigate's config. However, if you are running on a dedicated device that's separate from your proxy or if you expose Frigate directly to the internet, you may want to configure TLS with valid certificates.

@@ -45,7 +45,9 @@ frigate:
  ...
 ```

-Within the folder, the private key is expected to be named `privkey.pem` and the certificate is expected to be named `fullchain.pem`.
+Within the folder, the private key is expected to be named `privkey.pem` and the certificate is expected to be named `fullchain.pem`. Mounted certificates take precedence over the self signed pair in `/config/tls`.
+
+`privkey.pem` must be readable by the runtime user that runs NGINX. Frigate hands it over at startup when the mount is writable; on a `:ro` mount, make it readable by uid 1000 (or your `PUID`) yourself. See [Running as a non-root user](/configuration/non_root).

 Note that certbot uses symlinks, and those can't be followed by the container unless it has access to the targets as well, so if using certbot you'll also have to mount the `archive` folder for your domain, e.g.:

@@ -59,7 +61,7 @@ frigate:

 ```

-Frigate automatically compares the fingerprint of the certificate at `/etc/letsencrypt/live/frigate/fullchain.pem` against the fingerprint of the TLS cert in NGINX every minute. If these differ, the NGINX config is reloaded to pick up the updated certificate.
+Frigate automatically compares the fingerprint of the certificate it loaded, from either location, against the fingerprint of the TLS cert in NGINX every minute. If these differ, the NGINX config is reloaded to pick up the updated certificate.

 If you issue Frigate valid certificates you will likely want to configure it to run on port 443 so you can access it without a port number like `https://your-frigate-domain.com` by mapping 8971 to 443.

@@ -73,4 +75,4 @@ frigate:

 ## ACME Challenge

-Frigate also supports hosting the acme challenge files for the HTTP challenge method if needed. The challenge files should be mounted at `/etc/letsencrypt/www`.
+Frigate also supports hosting the acme challenge files for the HTTP challenge method if needed. The challenge files should be mounted at `/etc/letsencrypt/www`. With a read-only root filesystem this has to be a mounted volume, since Frigate cannot create the directory itself.
@@ -204,11 +204,20 @@ Light guidelines and advice:
 npm run lint
 ```

- Add to unit tests and ensure they pass. As much as possible, you should strive to _increase_ test coverage whenever making changes. This will help ensure features do not accidentally become broken in the future.
- If you run into error messages like "TypeError: Cannot read properties of undefined (reading 'context')" when running tests, this may be due to these issues (https://github.com/vitest-dev/vitest/issues/1910, https://github.com/vitest-dev/vitest/issues/1652) in vitest, but I haven't been able to resolve them.
+- Ensure the backend [unit tests](#unit-tests) pass. Your PR cannot be merged unless tests pass.
+
+```shell
+python3 -u -m unittest
+```
+
+- Ensure the end-to-end tests pass. They run in Playwright against a production build with mocked API data, so they don't need a running Frigate instance. Add or update tests in `web/e2e/specs/` when you change UI behavior.

 ```console
-npm run test
+# First-time setup
+npx playwright install chromium
+
+# Build the app and run all tests
+npm run e2e:build && npm run e2e
 ```

 - Test in different browsers. Firefox, Chrome, and Safari all have different quirks that make them unique targets to interact with.
@@ -54,7 +54,7 @@ Frigate supports multiple different detectors that work on different types of ha

 **Most Hardware**

- [Hailo](#hailo-8): The Hailo8 and Hailo8L AI Acceleration module is available in m.2 format with a HAT for RPi devices offering a wide range of compatibility with devices.
+- [Hailo](#hailo-8): The Hailo-8, Hailo-8L and Hailo-8R AI Acceleration modules are available in m.2 format with a HAT for RPi devices offering a wide range of compatibility with devices.
  - [Supports many model architectures](../../configuration/object_detectors#configuration-hailo)
  - Runs best with tiny or small size models

@@ -65,6 +65,11 @@ Frigate supports multiple different detectors that work on different types of ha
  - [Supports many model architectures](../../configuration/object_detectors#memryx-mx3)
  - Runs best with tiny, small, or medium-size models

+- <CommunityBadge /> [DEEPX](#deepx-npu): The DEEPX NPU is available in m.2 format and as a HAT+ for the Raspberry Pi 5, allowing for a wide range of compatibility with devices.
+  - [Supports YOLO model architectures](../../configuration/object_detectors#deepx-npu)
+  - Runs best with tiny or small size models
+  - Runs efficiently on low power hardware
+
 **AMD**

 - [ROCm](#rocm---amd-gpu): ROCm can run on AMD Discrete GPUs to provide efficient object detection
@@ -111,12 +116,13 @@ Frigate supports multiple different detectors that work on different types of ha

 ### Hailo-8

-Frigate supports both the Hailo-8 and Hailo-8L AI Acceleration Modules on compatible hardware platforms, including the Raspberry Pi 5 with the PCIe hat from the AI kit. The Hailo detector integration in Frigate automatically identifies your hardware type and selects the appropriate default model when a custom model isn’t provided.
+Frigate supports the Hailo-8, Hailo-8L and Hailo-8R AI Acceleration Modules on compatible hardware platforms, including the Raspberry Pi 5 with the PCIe hat from the AI kit. The Hailo detector integration in Frigate identifies which of them is attached and selects the matching default model when a custom model isn’t provided.

 **Default Model Configuration:**

- **Hailo-8L:** Default model is **YOLOv6n**.
- **Hailo-8:** Default model is **YOLOv6n**.
+- **Hailo-8L:** Default model is **YOLOv6n**, compiled for the Hailo-8L.
+- **Hailo-8:** Default model is **YOLOv6n**, compiled for the Hailo-8.
+- **Hailo-8R:** Default model is the **Hailo-8** build of **YOLOv6n**, since the Hailo Model Zoo publishes no Hailo-8R build.

 In real-world deployments, even with multiple cameras running concurrently, Frigate has demonstrated consistent performance. Testing on x86 platforms, with dual PCIe lanes, yields further improvements in FPS, throughput, and latency compared to the Raspberry Pi setup.

@@ -256,6 +262,32 @@ The MX3 is a pipelined architecture, where the maximum frames per second support

 Inference speeds may vary depending on the host platform. The above data was measured on an **Intel 13700 CPU**. Platforms like Raspberry Pi, Orange Pi, and other ARM-based SBCs have different levels of processing capability, which may limit total FPS.

+### DEEPX NPU
+
+Frigate supports the DEEPX NPU in both of its form factors: the **DX-M1** M.2 module, which works on x86 (Intel/AMD) and ARM-based SBCs such as the Raspberry Pi 5, and the **DX-M1M** on the [Sixfab AI HAT+](https://docs.sixfab.com/docs/ai-hat-plus-raspberry-pi-5-quickstart) for the Raspberry Pi 5. Both use the same driver and runtime, so the configuration is identical for either one. DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
+
+The DEEPX driver and runtime run on the Docker host rather than inside the Frigate container and must be installed before the NPU can be used. See the [installation docs](installation.md#deepx-npu) for the setup steps and [the detector docs](/configuration/object_detectors#deepx-npu) for the configuration.
+
+Frigate does not bundle a model for this detector. Models use DEEPX's `.dxnn` format, and pre-compiled YOLO models can be downloaded from the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo). Prefer a model with a `_ppu` suffix whenever one is available for the architecture you want: these run part of the post-processing on the NPU itself and are considerably faster, roughly 2.5x for the same architecture and input size. **YOLOX-S with PPU is the recommended starting point.**
+
+Inference times for a few recommended models, measured through Frigate's own stats on a DX-M1:
+
+| Model             | Input Size | DX-M1 Inference Time |
+| ----------------- | ---------- | -------------------- |
+| YOLOX-S (PPU)     | 640        | ~ 13 ms              |
+| YOLOv9-t (PPU)    | 640        | ~ 18 ms              |
+| YOLOv4 (PPU)      | 512        | ~ 20 ms              |
+| YOLOX-S           | 640        | ~ 34 ms              |
+| YOLOv9-s          | 640        | ~ 39 ms              |
+
+Other ModelZoo YOLO variants are also supported but have not been measured. Inference speeds vary with the host platform, so a slower host such as a Raspberry Pi 5 will report higher times than those above.
+
+:::note
+
+A few ModelZoo models can not be used with Frigate: SSD models (they are trained on Pascal VOC, so their labels do not match Frigate's), DAMO-YOLO models, face and pose models, and the PPU builds of YOLOv7.
+
+:::
+
 ### Nvidia Jetson

 Jetson devices are supported via the TensorRT or ONNX detectors when running Jetpack 6. It will [make use of the Jetson's hardware media engine](/configuration/hardware_acceleration_video#nvidia-jetson) when configured with the [appropriate presets](/configuration/ffmpeg_presets#hwaccel-presets), and will make use of the Jetson's GPU and DLA for object detection when configured with the [TensorRT detector](/configuration/object_detectors#nvidia-tensorrt-detector).
@@ -122,7 +122,9 @@ Additionally, the USB Coral draws a considerable amount of power. If using any o

 ### Hailo-8

-The Hailo-8 and Hailo-8L AI accelerators are available in both M.2 and HAT form factors for the Raspberry Pi. The M.2 version typically connects to a carrier board for PCIe, which then interfaces with the Raspberry Pi 5 as part of the AI Kit. The HAT version can be mounted directly onto compatible Raspberry Pi models. Both form factors have been successfully tested on x86 platforms as well, making them versatile options for various computing environments.
+The Hailo-8, Hailo-8L and Hailo-8R AI accelerators are available in both M.2 and HAT form factors for the Raspberry Pi. The M.2 version typically connects to a carrier board for PCIe, which then interfaces with the Raspberry Pi 5 as part of the AI Kit. The HAT version can be mounted directly onto compatible Raspberry Pi models. Both form factors have been successfully tested on x86 platforms as well, making them versatile options for various computing environments.
+
+The HailoRT runtime is not part of the Frigate image; Frigate downloads and installs it at first start once a Hailo detector is configured. Containers without internet access can provide the files themselves, see [Detector runtimes](/frigate/network_requirements#detector-runtimes).

 #### Installation

@@ -298,7 +300,7 @@ If you are using `docker run`, add this option to your command `--device /dev/ha

 #### Configuration

-Finally, configure [hardware object detection](/configuration/object_detectors#hailo-8) to complete the setup.
+Finally, configure [hardware object detection](/configuration/object_detectors#hailo) to complete the setup.

 ### MemryX MX3

@@ -315,6 +317,8 @@ The MemryX MX3 Accelerator is available in the M.2 2280 form factor (like an NVM

 To get started with MX3 hardware setup for your system, refer to the [Hardware Setup Guide](https://developer.memryx.com/2p1/get_started/install_hardware.html).

+The MemryX SDK used inside the container is not part of the Frigate image; Frigate downloads and installs it at first start once a MemryX detector is configured. Containers without internet access can provide the file themselves, see [Detector runtimes](/frigate/network_requirements#detector-runtimes). The host side driver still has to be installed as described below.
+
 Then follow these steps for installing the correct driver/runtime configuration:

 1. Copy or download [this script](https://github.com/blakeblackshear/frigate/blob/dev/docker/memryx/user_installation.sh).
@@ -377,6 +381,99 @@ If you can't use Docker Compose, you can run the container with something simila

 Finally, configure [hardware object detection](/configuration/object_detectors#memryx-mx3) to complete the setup.

+### DEEPX NPU
+
+The DEEPX NPU is available in two form factors, and Frigate supports both:
+
+- **DX-M1** in the M.2 2280 form factor (like an NVMe SSD), for x86 (Intel/AMD) PCs, the Raspberry Pi 5, and other ARM SBCs with an exposed PCIe M.2 slot.
+- **DX-M1M** on the [Sixfab AI HAT+](https://docs.sixfab.com/docs/ai-hat-plus-raspberry-pi-5-quickstart), a HAT+ board that connects to the Raspberry Pi 5 over PCIe Gen 3 x1.
+
+Both present the NPU through the same PCIe driver and DX-RT runtime, so the setup below and the detector configuration are identical for either one. Nothing needs to change when moving between them.
+
+DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
+
+#### Versions
+
+A DEEPX install has several separately versioned pieces, and they all have to agree. The driver, the runtime, and the daemon live on the Docker host; Frigate itself carries only the Python bindings, which it downloads on first start:
+
+| Component      | Version  | Installed on | Installed by              |
+| -------------- | -------- | ------------ | ------------------------- |
+| Kernel driver  | `v2.6.0` | Host  | `user_installation.sh`    |
+| DX-RT runtime  | `v3.4.0` | Host  | `user_installation.sh`    |
+| NPU firmware   | `v2.7.4` | The module   | Flashed from the host     |
+| DX-RT bindings | `v3.4.0` | Frigate      | Downloaded at first start |
+
+:::warning
+
+A version mismatch does not produce a startup error. It typically shows up as inference requests that are accepted but never return a result, so detections simply stop appearing while Frigate looks healthy. If that happens after a Frigate upgrade, check every version in the table before anything else.
+
+:::
+
+The installation script installs the DX-RT runtime on the host and enables `dxrt.service`, so the daemon starts at boot and any other program on the host can share the NPU with Frigate. Check the firmware version with `dxrt-cli --status` and update the module if it does not match the table above.
+
+#### Installation
+
+The DEEPX kernel driver must be installed on the host rather than in the container, because containers share the host kernel and cannot load kernel modules. Installing it creates the `/dev/dxrt*` device nodes that are passed through to Frigate. The same script installs the DX-RT runtime and enables `dxrt.service`, the daemon that owns the NPU and hands work to it on behalf of Frigate and anything else on the host.
+
+1. Copy or download [this script](https://github.com/blakeblackshear/frigate/blob/dev/docker/deepx/user_installation.sh).
+2. Ensure it has execution permissions with `sudo chmod +x user_installation.sh`
+3. Run the script with `./user_installation.sh`
+4. **Restart your computer** to complete driver installation.
+
+Confirm the NPU is visible before continuing:
+
+```bash
+ls /dev/dxrt*
+```
+
+Then confirm the daemon is running and listening in `/run/dxrt`:
+
+```bash
+systemctl is-active dxrt.service
+ls /run/dxrt/
+```
+
+#### Setup
+
+To set up Frigate, follow the default installation instructions, for example: `ghcr.io/blakeblackshear/frigate:stable`
+
+#### Docker configuration
+
+Frigate needs the NPU device node and the directory holding the daemon's socket:
+
+```yaml
+services:
+  frigate:
+    devices:
+      - /dev/dxrt0:/dev/dxrt0
+    volumes:
+      - /run/dxrt:/run/dxrt
+```
+
+If you can't use Docker Compose, add `--device /dev/dxrt0:/dev/dxrt0 -v /run/dxrt:/run/dxrt` to your `docker run` command.
+
+Add one `--device` per NPU, contiguously from `/dev/dxrt0`, since the client stops enumerating at the first gap.
+
+The installation script configures `dxrt.service` to place its socket in `/run/dxrt` through a systemd drop-in. Mounting the directory rather than the socket file means the container sees the new socket after `dxrt.service` is restarted, rather than holding on to a deleted one.
+
+`dxrtd` listens on an abstract socket as well, but that one does not cross into a container, so Frigate names the filesystem socket through `DXRT_DYNAMIC_IPC_ENDPOINT` on your behalf. Set that variable on the container yourself only if the daemon listens somewhere else, which means you also set it for `dxrtd` through its own systemd drop-in. The script writes `/etc/systemd/system/dxrt.service.d/frigate.conf` for exactly that, and has `dxrt.service` link the socket to `/tmp/dxrt_dynamic_ipc.sock` when it starts, so the host's own `dxrt-cli` and `dxtop` keep finding it at the default path they fall back to.
+
+:::note
+
+The DX-RT client exits when `dxrt.service` stops, so restart the Frigate container after restarting `dxrt.service`.
+
+:::
+
+The device node is needed as well as the socket, because the client opens the NPU directly even though the daemon arbitrates access. Without it, inference fails with `Device not found`.
+
+`/dev/shm` does not need sharing.
+
+The DX-RT python bindings are not shipped in the Frigate image. Frigate downloads them on first start when a DEEPX detector is configured, and caches them under `/config`.
+
+#### Configuration
+
+Finally, configure [hardware object detection](/configuration/object_detectors#deepx-npu) to complete the setup.
+
 ### Rockchip platform

 Make sure that you use a linux distribution that comes with the rockchip BSP kernel 5.10 or 6.1 and necessary drivers (especially rkvdec2 and rknpu). To check, enter the following commands:
@@ -479,6 +576,8 @@ Follow these steps for installation:

 To set up Frigate, follow the default installation instructions, for example: `ghcr.io/blakeblackshear/frigate:stable`

+The AXEngine python package is not part of the Frigate image; Frigate downloads and installs it at first start once an AXEngine detector is configured. Containers without internet access can provide the file themselves, see [Detector runtimes](/frigate/network_requirements#detector-runtimes).
+
 Next, grant Docker permissions to access your hardware by adding the following lines to your `docker-compose.yml` file:

 ```yaml
@@ -514,7 +613,7 @@ Generate a Frigate Docker Compose configuration based on your hardware and requi
 services:
  frigate:
    container_name: frigate
-    privileged: true # this may not be necessary for all setups
+    # privileged: true # ONLY enable if your hardware requires it (see hardware-specific docs); prefer the device mappings below
    restart: unless-stopped
    stop_grace_period: 30s # allow enough time to shut down the various services
    image: ghcr.io/blakeblackshear/frigate:stable
@@ -546,6 +645,30 @@ services:
  </TabItem>
 </Tabs>

+### Recommended security options
+
+Frigate does not need elevated container privileges for most setups. The following hardens the container; add the `devices`/`group_add` entries your hardware requires (see the hardware acceleration docs):
+
+```yaml
+services:
+  frigate:
+    ...
+    security_opt:
+      - no-new-privileges:true
+    cap_drop:
+      - ALL
+```
+
+:::note
+
+`telemetry.stats.network_bandwidth` uses nethogs, which requires root with NET_ADMIN/NET_RAW capabilities. If you enable that stat, omit `cap_drop: [ALL]` or add `cap_add: [NET_ADMIN, NET_RAW]`.
+
+Platforms that genuinely require `privileged: true` (MemryX, some QNAP setups) are called out in their own sections and are unaffected by this guidance.
+
+:::
+
+Frigate's services run as an unprivileged user inside the container. See [Running as a non-root user](../configuration/non_root.md) for the run modes, the one time volume ownership migration, what each accelerator needs on the host, and the [hardened deployment](../configuration/non_root.md#hardened-deployment) layout with a read-only root filesystem.
+
 **Docker CLI**

 If you can't use Docker Compose, you can run the container with something similar to this:
@@ -612,6 +735,8 @@ Home Assistant OS users can install via the App repository.
 5. Start the App
 6. Use the _Open Web UI_ button to access the Frigate UI, then click in the _cog icon_ > _Configuration editor_ and configure Frigate to your liking

+App users who can't set container environment variables can put `FRIGATE_` values in a `secrets.yaml` next to `config.yml` in `/addon_configs/<addon_directory>` instead. See [`secrets.yaml`](../configuration/advanced/system.md#secretsyaml).
+
 There are several variants of the App available:

 | App Variant                | Description                                                |
@@ -34,6 +34,12 @@ The following models are downloaded automatically the first time their associate
 | [Custom classification](/configuration/custom_classification/state_classification) (training) | MobileNetV2 ImageNet base weights (via Keras)                              | Google storage       |
 | [Audio transcription](/configuration/advanced/system)                                         | Whisper or Sherpa-ONNX streaming model                                     | HuggingFace / OpenAI |

+:::note
+
+The MobileNetV2 base weights are the one exception to the `/config/model_cache/` rule. They are also the only entry that is not downloaded when the feature is enabled: Frigate fetches them when a training run actually starts.
+
+:::
+
 ### Hardware-Specific Detector Models

 If you are using one of the following hardware detectors and have not provided your own model file, a default model will be downloaded on first startup:
@@ -41,7 +47,7 @@ If you are using one of the following hardware detectors and have not provided y
 | Detector                                                           | Model Downloaded     | Source                   |
 | ------------------------------------------------------------------ | -------------------- | ------------------------ |
 | [Rockchip RKNN](/configuration/object_detectors#rockchip-platform) | RKNN detection model | GitHub                   |
-| [Hailo 8 / 8L](/configuration/object_detectors#hailo-8)            | YOLOv6n (.hef)       | Hailo Model Zoo (AWS S3) |
+| [Hailo 8 / 8L / 8R](/configuration/object_detectors#hailo)         | YOLOv6n (.hef)       | Hailo Model Zoo (AWS S3) |
 | [AXERA AXEngine](/configuration/object_detectors)                  | Detection model      | HuggingFace              |

 :::note
@@ -50,6 +56,24 @@ The default CPU, EdgeTPU, and OpenVINO object detection models are bundled into

 :::

+### Detector Runtimes
+
+The SDKs for a few hardware detectors are not shipped in the Frigate image. They are downloaded the first time that detector is configured, verified against checksums pinned in the Frigate release, and installed into the Frigate user's home directory (`/config/.local` by default). Once installed they are not downloaded again until a Frigate release pins a new version.
+
+| Detector                                                   | Version | Files                                                                                                                                                                                 | Source                                                                                   |
+| ---------------------------------------------------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
+| [Hailo 8 / 8L / 8R](/configuration/object_detectors#hailo) | 4.21.0  | `hailort-debian12-amd64.tar.gz` and `hailort-4.21.0-cp311-cp311-linux_x86_64.whl` on x86, `hailort-debian12-arm64.tar.gz` and `hailort-4.21.0-cp311-cp311-linux_aarch64.whl` on arm64 | [GitHub release](https://github.com/frigate-nvr/hailort/releases/tag/v4.21.0)            |
+| [MemryX MX3](/configuration/object_detectors#memryx-mx3)   | 2.1.0   | `mx_accl_frigate-2.1.0.zip` (the release source archive, renamed)                                                                                                                     | [GitHub archive](https://github.com/memryx/mx_accl_frigate/archive/refs/tags/v2.1.0.zip) |
+| [AXERA AXEngine](/configuration/object_detectors#axera)    | 0.1.3   | `axengine-0.1.3-py3-none-any.whl`                                                                                                                                                     | [GitHub release](https://github.com/AXERA-TECH/pyaxengine/releases/tag/0.1.3-frigate)    |
+
+If the container cannot reach GitHub, provide the files yourself:
+
+1. Download the files for your architecture on a machine with internet access.
+2. Place them, with exactly the file names listed above, in `/config/model_cache/runtimes/<detector>/`, where `<detector>` is the detector named in your config's `devices` (`hailo`, `memryx`, or `axengine`).
+3. Start Frigate. Files whose checksum matches are installed without any download; a file with the wrong checksum is discarded and downloaded again, so a failed startup log names the file to replace.
+
+The `GITHUB_ENDPOINT` mirror variable below applies to these downloads as well.
+
 ### Preventing Model Downloads

 If you have already downloaded all required models and want to prevent Frigate from attempting any outbound connections to HuggingFace or the Transformers library, set the following environment variables on your Frigate container:
@@ -73,9 +97,9 @@ If your Frigate instance has restricted internet access, you can point model dow
 | Environment Variable                | Default                             | Used By                                       |
 | ----------------------------------- | ----------------------------------- | --------------------------------------------- |
 | `HF_ENDPOINT`                       | `https://huggingface.co`            | Semantic search, Sherpa-ONNX, AXEngine models |
-| `GITHUB_ENDPOINT`                   | `https://github.com`                | Face recognition, LPR, RKNN models            |
+| `GITHUB_ENDPOINT`                   | `https://github.com`                | Face recognition, LPR, RKNN models, detector runtimes |
 | `GITHUB_RAW_ENDPOINT`               | `https://raw.githubusercontent.com` | Bird classification                           |
-| `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` | Google storage (Keras default)      | Custom classification training                |
+| `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` | Unset (Keras uses its own default)  | Custom classification training                |

 ## Optional Cloud Services

@@ -118,16 +142,12 @@ When [notifications](/configuration/notifications) are enabled and users have re

 If an [MQTT broker](/integrations/mqtt) is configured, Frigate maintains a connection to the broker's host and port. This is typically a local network connection, but will require internet if you use a cloud-hosted MQTT broker.

-### DeepStack / CodeProject.AI
-
-When using the [DeepStack detector plugin](/configuration/object_detectors), Frigate sends images to the configured API endpoint for inference. This is typically local but depends on where the service is hosted.
-
 ## WebRTC (STUN)

 For [WebRTC live streaming](/configuration/live), Frigate uses STUN for NAT traversal:

 - **go2rtc** defaults to a local STUN listener (`stun:8555`), no internet required.
- **The web UI's WebRTC player** includes a fallback to Google's public STUN server (`stun:stun.l.google.com:19302`), which requires internet.
+- **The web UI** uses the servers in `go2rtc.webrtc.ice_servers` for its WebRTC player and for the WebRTC connectivity check it runs when the Live view loads. If none are set, it uses Google's public STUN server (`stun:stun.l.google.com:19302`), which requires internet access from the browser. Set `ice_servers` to a STUN or TURN server on your network to avoid this.

 ## Home Assistant Supervisor

@@ -147,9 +167,23 @@ When running as a Home Assistant App, the go2rtc startup script queries the loca
 To run Frigate in an air-gapped or offline environment:

 1. **Pre-download models**: Start Frigate with internet access once with all desired features enabled. Models will be cached in `/config/model_cache/`.
-2. **Disable version check**: Set `telemetry.version_check: false` in your configuration.
-3. **Block outbound model requests**: Set the `HF_HUB_OFFLINE=1` and `TRANSFORMERS_OFFLINE=1` environment variables to prevent HuggingFace and Transformers from attempting any network requests.
-4. **Avoid cloud features**: Do not configure Frigate+, Generative AI providers that require internet, or cloud MQTT brokers.
-5. **Use local model mirrors**: If limited internet is available, set the `HF_ENDPOINT`, `GITHUB_ENDPOINT`, and `GITHUB_RAW_ENDPOINT` environment variables to point to local mirrors.
+2. **Pre-download the training base weights**: If you plan to train custom classification models, set `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` before training, then run one training job while online. Without this variable the base weights are cached outside `/config/` and are lost whenever the container is recreated, so a later training run will fail offline. If the machine never has internet access, copy the weights in manually as described below.
+3. **Disable version check**: Set `telemetry.version_check: false` in your configuration.
+4. **Block outbound model requests**: Set the `HF_HUB_OFFLINE=1` and `TRANSFORMERS_OFFLINE=1` environment variables to prevent HuggingFace and Transformers from attempting any network requests.
+5. **Avoid cloud features**: Do not configure Frigate+, Generative AI providers that require internet, or cloud MQTT brokers.
+6. **Use local model mirrors**: If limited internet is available, set the `HF_ENDPOINT`, `GITHUB_ENDPOINT`, `GITHUB_RAW_ENDPOINT`, and `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` environment variables to point to local mirrors.

 After these steps, Frigate will operate with no outbound internet connections.
+
+### Manually Copying the Training Base Weights
+
+On a machine with internet access, download the weights:
+
+```bash
+curl -L -o mobilenet_v2_weights.h5 \
+  "https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_224_no_top.h5"
+```
+
+Copy the file into your Frigate config volume as `/config/model_cache/MobileNet/mobilenet_v2_weights.h5`, keeping that exact filename, then set the environment variable `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` in your Docker compose file to the URL above and restart Frigate.
+
+The variable must be set even though the URL is never contacted. If it is unset, Frigate ignores the copied file and asks Keras to download the weights instead.
@@ -60,7 +60,7 @@ If you’re running Frigate via Docker (recommended method), follow these steps:
     ```bash
     docker logs frigate
     ```
-   - Visit the Frigate Web UI (default: `http://<your-ip>:5000`) to confirm the new version is running. The version number is displayed at the top of the System Metrics page.
+   - Visit the Frigate Web UI (default: `http://<your-ip>:5000`) to confirm the new version is running. The version number is displayed at the top of the Health and Metrics page.

 ### Notes

@@ -4,6 +4,7 @@ title: Getting started
 ---

 import ConfigTabs from "@site/src/components/ConfigTabs";
+import Tabs from "@theme/Tabs";
 import TabItem from "@theme/TabItem";
 import NavPath from "@site/src/components/NavPath";

@@ -132,21 +133,68 @@ services:
      - "8554:8554" # RTSP feeds
 ```

-Now you should be able to start Frigate by running `docker compose up -d` from within the folder containing `docker-compose.yml`. On startup, an admin user and password will be created and outputted in the logs. You can see this by running `docker logs frigate`. Frigate should now be accessible at `https://server_ip:8971` where you can login with the `admin` user and finish configuration using the Settings UI.
+Now you should be able to start Frigate by running `docker compose up -d` from within the folder containing `docker-compose.yml`. On startup, an admin user and password will be created and outputted in the logs. You can see this by running `docker logs frigate`. Frigate should now be accessible at `https://server_ip:8971` where you can login with the `admin` user. With no cameras configured yet, the setup wizard runs on first login and walks you through the rest.

 ## Configuring Frigate

 This section assumes that you already have an environment setup as described in [Installation](../frigate/installation.md). You should also configure your cameras according to the [camera setup guide](/frigate/camera_setup). Pay particular attention to the section on choosing a detect resolution.

-### Step 1: Start Frigate
+<Tabs
+groupId="setup-method"
+defaultValue="wizard"
+values={[
+{ label: "Setup wizard", value: "wizard" },
+{ label: "Manual", value: "manual" },
+]}
+
+> <TabItem value="wizard">
+
+The first time you open Frigate with no cameras configured, the setup wizard walks you through the basics. Every step can be skipped, everything it sets can be changed later in Settings, and once you finish or dismiss it, it doesn't come back.
+
+:::note
+
+Frigate only sees hardware that has been passed into the container. If you plan to use a GPU, a Coral, or another accelerator, add the device to your `docker-compose.yml` and restart before running the wizard, otherwise it won't appear in the detection or hardware acceleration steps. The Manual tab shows the device entries for an Intel or AMD GPU and for a Coral, and the [hardware acceleration](../configuration/hardware_acceleration_video.md) and [object detectors](../configuration/object_detectors.md) docs cover the rest.
+
+:::
+
+**Account**
+
+Set a password for the `admin` account to replace the generated one from the logs, and add accounts for anyone else who needs access. This step is hidden if you have turned authentication off.
+
+**Add a camera**
+
+Opens the [Add Camera Wizard](../configuration/cameras.md#adding-a-camera-with-the-add-camera-wizard), which connects to the camera, tests each stream, and writes its configuration for you. You can add more than one before moving on.
+
+**Object detection**
+
+Lists the detection hardware Frigate found on your system, such as a Coral, an Intel GPU or NPU, or a discrete GPU, and configures the one you pick. NVIDIA and AMD GPUs need a model before detection can start, so the wizard offers your Frigate+ models if you have them, or lets you finish setup and add one later under <NavPath path="Settings > System > Detection models" />.
+
+**Hardware acceleration**
+
+Offers only the decoding methods your hardware supports. Auto picks one based on that hardware and the codec your camera sends, so a mixed h264 and h265 setup gets the right preset per camera.
+
+**Recording**
+
+Choose whether to record only when something is detected or around the clock, and how long to keep it.
+
+The last screen summarizes what was set up. If a step changed something that needs a restart, the button restarts Frigate and returns you to the Live view once it is back.
+
+The wizard configures the essentials only. Motion masks are not included and should be set up afterward, once you can identify the areas of the frame that trigger unwanted motion. See the [masks documentation](../configuration/masks.md). Zones, tracked object types, notifications, and MQTT are also configured in Settings.
+
+</TabItem>
+<TabItem value="manual">
+
+On a new install the setup wizard opens first. Click **Skip setup and configure manually** on its welcome screen to dismiss it, and the steps below apply. The wizard won't come back once dismissed.
+
+**Step 1: Start Frigate**

 At this point you should be able to start Frigate and a basic config will be created automatically.

-### Step 2: Add a camera
+**Step 2: Add a camera**

 Click the **Add Camera** button in <NavPath path="Settings > Global configuration > Camera management" /> to use the camera setup wizard to get your first camera added into Frigate. See [Adding a camera with the Add Camera Wizard](../configuration/cameras.md#adding-a-camera-with-the-add-camera-wizard) for a walkthrough of each step.

-### Step 3: Configure hardware acceleration (recommended)
+**Step 3: Configure hardware acceleration (recommended)**

 Now that you have a working camera configuration, set up hardware acceleration to minimize the CPU required to decode your video streams. See the [hardware acceleration](../configuration/hardware_acceleration_video.md) docs for examples applicable to your hardware.

@@ -190,7 +238,7 @@ cameras:
 </TabItem>
 </ConfigTabs>

-### Step 4: Configure detectors
+**Step 4: Configure detectors**

 By default, Frigate will use a single OpenVINO detector running on the CPU.

@@ -204,8 +252,8 @@ You need to refer to **Configure hardware acceleration** above to enable the con
 <ConfigTabs>
 <TabItem value="ui">

-1. Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a detector with **Type** `OpenVINO` and **Device** `GPU`
-2. On the same page, in the **Custom Model** tab, configure the model settings for OpenVINO:
+1. Navigate to <NavPath path="Settings > System > Detection models" /> and select **Intel GPU** from the **Hardware** dropdown
+2. On the same model, open the **Custom Model** tab and configure the model settings for OpenVINO:

 | Field                                    | Value                                      |
 | ---------------------------------------- | ------------------------------------------ |
@@ -222,15 +270,12 @@ You need to refer to **Configure hardware acceleration** above to enable the con
 ```yaml {3-6,9-15,20-21}
 mqtt: ...

-detectors: # <---- add detectors
-  ov:
-    type: openvino  # <---- use openvino detector
-    device: GPU
-
-# We will use the default MobileNet_v2 model from OpenVINO.
-model:
-  width: 300
-  height: 300
+models: # <---- add models
+  - devices:
+      - openvino:GPU # <---- use the openvino detector on the GPU
+    # We will use the default MobileNet_v2 model from OpenVINO.
+    width: 300
+    height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  path: /openvino-model/ssdlite_mobilenet_v2.xml
@@ -273,7 +318,7 @@ services:
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a detector with **Type** `EdgeTPU` and **Device** `usb`.
+Navigate to <NavPath path="Settings > System > Detection models" /> and select **Coral EdgeTPU (USB)** from the **Hardware** dropdown.

 </TabItem>
 <TabItem value="yaml">
@@ -281,10 +326,9 @@ Navigate to <NavPath path="Settings > System > Detectors and model" /> and add a
 ```yaml {3-6,11-12}
 mqtt: ...

-detectors: # <---- add detectors
-  coral:
-    type: edgetpu
-    device: usb
+models: # <---- add models
+  - devices:
+      - edgetpu:usb

 cameras:
  name_of_your_camera:
@@ -303,7 +347,7 @@ More details on available detectors can be found [here](../configuration/object_

 Restart Frigate and you should start seeing detections for `person`. If you want to track other objects, they can be configured in <NavPath path="Settings > Global configuration > Objects" /> or via the [configuration file reference](../configuration/advanced/reference.md).

-### Step 5: Setup motion masks
+**Step 5: Setup motion masks**

 Now that you have optimized your configuration for decoding the video stream, you will want to check to see where to implement motion masks. Click on the camera from the main dashboard, then select the gear icon in the top right, enable the [Debug view](/usage/live#the-single-camera-view), and finally enable the switch for Motion Boxes. Watch for areas that continuously trigger unwanted motion to be detected. Common areas to mask include camera timestamps and trees that frequently blow in the wind. The goal is to avoid wasting object detection cycles looking at these areas.

@@ -321,10 +365,9 @@ If you are using YAML to configure Frigate instead of the UI, your configuration
 mqtt:
  enabled: False

-detectors:
-  coral:
-    type: edgetpu
-    device: usb
+models:
+  - devices:
+      - edgetpu:usb

 cameras:
  name_of_your_camera:
@@ -341,7 +384,7 @@ cameras:
          coordinates: "0,461,3,0,1919,0,1919,843,1699,492,1344,458,1346,336,973,317,869,375,866,432"
 ```

-### Step 6: Enable recordings
+**Step 6: Enable recordings**

 In order to review activity in the Frigate UI, recordings need to be enabled.

@@ -357,7 +400,7 @@ In order to review activity in the Frigate UI, recordings need to be enabled.
 ```yaml {16-17}
 mqtt: ...

-detectors: ...
+models: ...

 cameras:
  name_of_your_camera:
@@ -390,7 +433,10 @@ If you only plan to use Frigate for recording, it is still recommended to define

 By default, Frigate will retain video of all tracked objects for 10 days. The full set of options for recording can be found [here](../configuration/advanced/reference.md).

-### Step 7: Complete config
+</TabItem>
+</Tabs>
+
+### Complete config

 At this point you have a complete config with basic functionality.

@@ -3,35 +3,100 @@ id: homekit
 title: HomeKit
 ---

-Frigate cameras can be integrated with Apple HomeKit through go2rtc. This allows you to view your camera streams directly in the Apple Home app on your iOS, iPadOS, macOS, and tvOS devices.
+Frigate cameras can be exported to Apple HomeKit through go2rtc. Each exported camera appears as an accessory in the Apple Home app on your iOS, iPadOS, macOS, and tvOS devices.

 ## Overview

-HomeKit integration is handled entirely through go2rtc, which is embedded in Frigate. go2rtc provides the necessary HomeKit Accessory Protocol (HAP) server to expose your cameras to HomeKit.
+Exporting cameras is handled entirely through go2rtc, which is embedded in Frigate. go2rtc provides the necessary HomeKit Accessory Protocol (HAP) server, so your camera is published to HomeKit as an accessory in its own right.

-## Setup
+:::note

-All HomeKit configuration and pairing should be done through the **go2rtc WebUI**.
+This is the opposite of importing a HomeKit camera. go2rtc can also pair with an existing HomeKit camera (Aqara, Eve, Eufy, and similar) and use it as a stream source, which is what the `add` page of the go2rtc WebUI is for. That page discovers HomeKit accessories on your network and will not list your Frigate cameras. It is not used for exporting.

-### Accessing the go2rtc WebUI
-
-The go2rtc WebUI is available at:
-
-```
-http://<frigate_host>:1984
-```
-
-Replace `<frigate_host>` with the IP address or hostname of your Frigate server.
-
-### Pairing Cameras
-
-1. Navigate to the go2rtc WebUI at `http://<frigate_host>:1984`
-2. Use the `add` section to add a new camera to HomeKit
-3. Follow the on-screen instructions to generate pairing codes for your cameras
+:::

 ## Requirements

- Frigate must be accessible on your local network using host network_mode
- Your iOS device must be on the same network as Frigate
- Port 1984 must be accessible for the go2rtc WebUI
- For detailed go2rtc configuration options, refer to the [go2rtc documentation](https://github.com/AlexxIT/go2rtc)
+- Frigate must be running with `network_mode: host` so that HomeKit can discover your cameras over mDNS
+- Your Apple device must be on the same network as Frigate
+- Port 1984 must be accessible so you can reach the go2rtc WebUI
+
+HomeKit also places strict limits on the stream itself. go2rtc passes your stream through without resizing or re-encoding it, so the stream you export must already meet these requirements:
+
+- **Video:** H.264 at 1920x1080, 1280x720, or 320x240
+- **Audio:** Opus, mono, 16 kHz
+
+A camera's full resolution stream usually does not qualify. See [Exporting a compatible stream](#exporting-a-compatible-stream) below.
+
+## Configuration
+
+HomeKit settings are stored in `/config/go2rtc_homekit.yml`. This is a separate file from your Frigate config, because go2rtc needs to write your pairings back to it when you pair a device.
+
+Edit it using the go2rtc config editor, which writes to that file directly:
+
+```
+http://<frigate_host>:1984/editor.html
+```
+
+Replace `<frigate_host>` with the IP address or hostname of your Frigate server. The editor will be empty until you add a HomeKit section, since this file holds only your HomeKit settings and not the rest of your go2rtc config.
+
+:::warning
+
+Do not put the `homekit:` section in the `go2rtc:` section of your Frigate config.
+
+Frigate regenerates that config on every startup, so go2rtc cannot save your pairings to it. Pairing will appear to succeed and then fail after the next restart with `PairVerify with unknown client_id`. If the section exists in both places, your saved pairings are erased on every restart.
+
+:::
+
+Add an entry for each camera you want to export. The key must match the name of a go2rtc stream, and the pin must be 8 digits. This is the number the Home app calls the setup code:
+
+```yaml
+homekit:
+  front_door:
+    name: Front Door
+    pin: "12345678"
+```
+
+If the key does not match a go2rtc stream, go2rtc logs `[homekit] missing stream:` at startup and the camera will not appear in the Home app.
+
+:::note
+
+go2rtc derives each accessory's HomeKit identity from this key, so renaming it later means the camera appears as a new accessory and has to be paired again. Settle on the name before you pair.
+
+:::
+
+Frigate keeps only the `homekit:` section of this file when it starts, so do not store streams or other go2rtc settings in it.
+
+### Exporting a compatible stream
+
+If a camera's stream does not meet the requirements listed above, define a scaled restream in your Frigate config and point HomeKit at that stream instead of the original:
+
+```yaml
+go2rtc:
+  streams:
+    front_door:
+      - rtsp://user:password@192.168.1.50:554/stream
+    front_door_homekit:
+      - "ffmpeg:front_door#video=h264#width=1280#height=720#audio=opus/16000"
+```
+
+```yaml
+# /config/go2rtc_homekit.yml
+homekit:
+  front_door_homekit:
+    name: Front Door
+    pin: "12345678"
+```
+
+Add `#hardware=cuda`, `#hardware=vaapi`, or the appropriate value for your system to transcode using your GPU. Note that NVENC cannot encode H.264 wider than 4096 pixels, so very wide streams must be scaled down as shown above rather than only re-encoded.
+
+## Pairing Cameras
+
+1. Restart Frigate after adding the `homekit:` section
+2. In the Apple Home app, choose **Add Accessory**, then **More options** to enter a code manually
+3. Select your camera and enter the pin you configured as the setup code
+4. Confirm that a `pairings:` list now appears under the camera in `/config/go2rtc_homekit.yml`
+
+Pairings are saved back to that file automatically. If step 4 shows no `pairings:` list, check the Frigate log for `[homekit] can't save`, which means the `homekit:` section is missing from `/config/go2rtc_homekit.yml`.
+
+For detailed go2rtc configuration options, refer to the [go2rtc documentation](https://github.com/AlexxIT/go2rtc).
@@ -292,7 +292,9 @@ Topic with the currently active profile name. Published value is the profile nam

 ### `frigate/notifications/set`

-Topic to turn notifications on and off. Expected values are `ON` and `OFF`.
+Topic to turn notifications on and off for all cameras. Expected values are `ON` and `OFF`.
+
+Only available when notifications are enabled in the config. Not persisted across Frigate restarts.

 ### `frigate/notifications/state`

@@ -302,12 +304,14 @@ Topic with current state of notifications. Published values are `ON` and `OFF`.

 ### `frigate/<camera_name>/status/<role>`

-Publishes the current health status of each role that is enabled (`audio`, `detect`, `record`). Possible values are:
+Publishes the current health status of each role that is enabled (`audio`, `detect`, `record`, `record_sub`). `record_sub` is only published for cameras with [sub stream recording](/configuration/record#sub-stream-recording) enabled, and is tracked separately from `record` so a healthy main stream can't hide a stalled sub stream. Possible values are:

 - `online`: Stream is running and being processed
 - `offline`: Stream is offline and is being restarted
 - `disabled`: Camera is currently turned off (either at runtime via the `enabled/set` topic, or persistently via the configuration file). See [Camera state](/configuration/live#camera-state) for the distinction.

+These reflect the state of Frigate's process for that role, not the camera's reachability, so an unreachable camera alternates between `offline` and `online` as the watchdog restarts ffmpeg. Wait for the status to hold steady (for example with Home Assistant's `for:`) rather than acting on a single message.
+
 ### `frigate/<camera_name>/<object_name>`

 Publishes the count of objects for the camera for use as a sensor in Home Assistant.
@@ -390,6 +394,18 @@ Topic to turn audio detection for a camera on and off. Expected values are `ON`

 Topic with current state of audio detection for a camera. Published values are `ON` and `OFF`.

+### `frigate/<camera_name>/audio_transcription/set`
+
+Topic to turn [live audio transcription](/configuration/audio_detectors#live-transcription) for a camera on and off. Expected values are `ON` and `OFF`. Transcribed text is published to `frigate/<camera_name>/audio/transcription`.
+
+`ON` is ignored unless audio transcription is enabled in the config for the camera. Unlike the other camera toggles, this one is not persisted across Frigate restarts.
+
+**NOTE:** Requires audio detection and transcription to be enabled
+
+### `frigate/<camera_name>/audio_transcription/state`
+
+Topic with current state of live audio transcription for a camera. Published values are `ON` and `OFF`.
+
 ### `frigate/<camera_name>/recordings/set`

 Topic to turn recordings for a camera on and off. Expected values are `ON` and `OFF`. The change is persisted across Frigate restarts (see [Runtime toggle persistence](/configuration/live#runtime-toggle-persistence)).
@@ -537,35 +553,42 @@ must be enabled in the configuration.

 Topic with current state of Birdseye for a camera. Published values are `ON` and `OFF`.

-### `frigate/<camera_name>/birdseye_mode/set`
+### `frigate/<camera_name>/birdseye_modes/set`

-Topic to set Birdseye mode for a camera. Birdseye offers different modes to customize under which circumstances the camera is shown.
+Topic to set the Birdseye activity types for a camera. Send one uppercase activity type or combine multiple types with commas, for example `MOTION,ALERTS`.

-_Note: Changing the value from `CONTINUOUS` -> `MOTION | OBJECTS` will take up to 30 seconds for
+_Note: Changing the value from `CONTINUOUS` to non-continuous activity types will take up to 30 seconds for
 the camera to be removed from the view._

-| Command      | Description                                                       |
-| ------------ | ----------------------------------------------------------------- |
-| `CONTINUOUS` | Always included                                                   |
-| `MOTION`     | Show when detected motion within the last 30 seconds are included |
-| `OBJECTS`    | Shown if an active object tracked within the last 30 seconds      |
+| Command       | Description                                                      |
+| ------------- | ---------------------------------------------------------------- |
+| `CONTINUOUS`  | Always included                                                  |
+| `MOTION`      | Shown if motion was detected within the last 30 seconds          |
+| `ALL_OBJECTS` | Shown if a tracked object was present within the last 30 seconds |
+| `ALERTS`      | Shown while an alert review item is in progress                  |
+| `DETECTIONS`  | Shown while a detection review item is in progress               |
+| `NONE`        | Never included                                                   |

-### `frigate/<camera_name>/birdseye_mode/state`
+### `frigate/<camera_name>/birdseye_modes/state`

-Topic with current state of the Birdseye mode for a camera. Published values are `CONTINUOUS`, `MOTION`, `OBJECTS`.
+Topic with the current Birdseye activity types for a camera. Multiple enabled types are published as a comma-separated value in the order `CONTINUOUS`, `MOTION`, `ALL_OBJECTS`, `ALERTS`, `DETECTIONS`. `NONE` is published when no activity types are enabled.

 ### `frigate/<camera_name>/notifications/set`

-Topic to turn notifications on and off. Expected values are `ON` and `OFF`.
+Topic to turn notifications for a camera on and off. Expected values are `ON` and `OFF`.
+
+`ON` is ignored unless notifications are enabled in the config for the camera. This is not persisted across Frigate restarts. It is the same control the UI labels **Suspend until restart**.

 ### `frigate/<camera_name>/notifications/state`

-Topic with current state of notifications. Published values are `ON` and `OFF`.
+Topic with current state of notifications. Published values are `ON` and `OFF`. This is the authoritative topic for whether a camera will notify.

 ### `frigate/<camera_name>/notifications/suspend`

-Topic to suspend notifications for a certain number of minutes. Expected value is an integer.
+Topic to suspend notifications for a certain number of minutes. Expected value is an integer. Separate from `notifications/set`: it does not change `notifications/state`, and is ignored while notifications are off.

 ### `frigate/<camera_name>/notifications/suspended`

-Topic with timestamp that notifications are suspended until. Published value is a UNIX timestamp, or 0 if notifications are not suspended.
+Topic with timestamp that notifications are suspended until. Published value is a UNIX timestamp, or 0 if there is no timed suspension.
+
+`0` does not mean notifications are enabled: `notifications/set` `OFF` clears the timed suspension, so this publishes `0` while `notifications/state` is `OFF`.
@@ -59,13 +59,12 @@ You can view all of your submitted images at [https://plus.frigate.video](https:

 Once you have [requested your first model](../plus/first_model.md) and gotten your own model ID, it can be used with a special model path. No other information needs to be configured for Frigate+ models because it fetches the remaining config from Frigate+ automatically.

-You can either choose the new model from the <NavPath path="Settings > System > Detectors and model" /> pane in the Frigate UI (the **Frigate+ Model** tab), or manually set the model at the root level in your config:
+You can either choose the new model from the <NavPath path="Settings > System > Detection models" /> pane in the Frigate UI (on the **Frigate+** tab of the model you want to change), or set it on that model in your config:

 ```yaml
-detectors: ...
-
-model:
-  path: plus://<your_model_id>
+models:
+  - devices: ...
+    path: plus://<your_model_id>
 ```

 :::note
@@ -79,10 +78,11 @@ Models are downloaded into the `/config/model_cache` folder and only downloaded
 If needed, you can override the labelmap for Frigate+ models. This is not recommended as renaming labels will break the Submit to Frigate+ feature if the labels are not available in Frigate+.

 ```yaml
-model:
-  path: plus://<your_model_id>
-  labelmap:
-    3: animal
-    4: animal
-    5: animal
+models:
+  - devices: ...
+    path: plus://<your_model_id>
+    labelmap:
+      3: animal
+      4: animal
+      5: animal
 ```
@@ -30,16 +30,15 @@ Models available in Frigate+ can be used with a special model path. No other inf
 <ConfigTabs>
 <TabItem value="ui">

-Navigate to <NavPath path="Settings > System > Detectors and model" />. In the **Detection Model** section, choose the **Frigate+** tab. Select your new Frigate+ model from the **Available Frigate+ models** dropdown, then click **Save**. Restart Frigate to apply the change.
+Navigate to <NavPath path="Settings > System > Detection models" />. On the model you want to change, choose the **Frigate+** tab and select your new Frigate+ model from the **Available Frigate+ models** dropdown, then click **Save**. Restart Frigate to apply the change.

 </TabItem>
 <TabItem value="yaml">

 ```yaml
-detectors: ...
-
-model:
-  path: plus://<your_model_id>
+models:
+  - devices: ...
+    path: plus://<your_model_id>
 ```

 :::tip
@@ -41,7 +41,7 @@ Rockchip models are automatically converted as of 0.17. For 0.16, YOLOv9 onnx mo

 ## Supported detector types

-Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo8l`), and Rockchip (`rknn`) detectors.
+Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo`), and Rockchip (`rknn`) detectors.

 | Hardware                                                                         | Recommended Detector Type | Recommended Model Type |
 | -------------------------------------------------------------------------------- | ------------------------- | ---------------------- |
@@ -50,7 +50,7 @@ Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVi
 | [Intel](/configuration/object_detectors.md#openvino-detector)                    | `openvino`                | `yolov9`               |
 | [NVidia GPU](/configuration/object_detectors#onnx)                               | `onnx`                    | `yolov9`               |
 | [AMD ROCm GPU](/configuration/object_detectors#amdrocm-gpu-detector)             | `onnx`                    | `yolov9`               |
-| [Hailo8/Hailo8L/Hailo8R](/configuration/object_detectors#hailo-8)                | `hailo8l`                 | `yolov9`               |
+| [Hailo8/Hailo8L/Hailo8R](/configuration/object_detectors#hailo)                  | `hailo`                   | `yolov9`               |
 | [Rockchip NPU](/configuration/object_detectors#rockchip-platform)                | `rknn`                    | `yolov9`               |

 ## Improving your model
@@ -34,11 +34,15 @@ The detect FFmpeg process exited on its own. This message is only the notificati

 </FaqItem>

-<FaqItem id="non-monotonically-increasing-dts" question="Application provided invalid, non monotonically increasing dts to muxer">
+<FaqItem id="non-monotonically-increasing-dts" question="Non-monotonic DTS / non monotonically increasing dts to muxer / Queue input is backward in time">

-An FFmpeg message meaning the camera sent packets with out-of-order timestamps. Because recordings are copied without re-encoding, FFmpeg cannot fix them, and the segment muxer often splits early, producing one-second segments and a cache backlog. The usual cause is a camera "Smart Codec" / H.264+ / H.265+ mode or a camera clock that jumps.
+These are FFmpeg messages indicating the camera sent packets with out-of-order timestamps, either on the video or the audio stream. Timestamp jitter like this is common with WiFi cameras and restreamed or proxied sources; other causes are a camera "Smart Codec" / H.264+ / H.265+ mode or a camera clock that jumps. A sustained flood of these messages usually precedes the stream stalling and the watchdog restarting FFmpeg.

-See [Recordings: segments are only 1 second long](/troubleshooting/recordings#segments-are-only-1-second-long).
+In most cases, the fix is to improve the network, reduce system resource usage, or switch to non-WiFi cameras. In general, WiFi cameras are [not recommended](https://ipcamtalk.com/threads/multiple-cameras-high-bandwidth.77100/#post-861110).
+
+On the video stream, this can affect recordings: because they are copied without re-encoding, FFmpeg cannot fix the timestamps, and the segment muxer often splits early, producing one-second segments and a cache backlog. See [Recordings: segments are only 1 second long](/troubleshooting/recordings#segments-are-only-1-second-long).
+
+On the audio stream, the messages can come from the output's audio encoding. If the audio stream is the problem, it may help to have go2rtc transcode it by adding `#audio=aac` to the camera's go2rtc stream to produce clean timestamps for everything consuming the restream.

 </FaqItem>

@@ -62,17 +66,19 @@ An FFmpeg message meaning it probed the stream but never saw enough decodable vi

 ## Recording

-<FaqItem id="no-new-recording-segments" question="No new recording segments were created for <camera> in the last 120s">
+<FaqItem id="no-new-recording-segments" question="No new recording segments were created (or: No new valid recording segments were created / No valid segments created since last invalid segment) for <camera> in the last 120s">

-Frigate's record watchdog is restarting the record FFmpeg process because no valid segment has reached the cache. This means the record stream is not connecting or the segments are being rejected (see the audio-codec entry below).
+Frigate's record watchdog is restarting the record FFmpeg process because the camera stopped producing usable recordings. The wording distinguishes the cases: `No new recording segments` means no new segment file reached the cache, so ffmpeg isn't getting video out of the record stream; the two `valid` variants mean recordings are arriving but keep failing validation. Either way the fault is on the camera or network side, and the restart is Frigate trying to recover.

-See [Recordings: the record stream isn't connecting](/troubleshooting/recordings#the-record-stream-isnt-connecting).
+See [Recordings: no new recording segments were created](/troubleshooting/recordings#no-new-recording-segments-were-created).

 </FaqItem>

-<FaqItem id="invalid-or-missing-video-stream-in-segment" question="Invalid or missing video stream in segment. Discarding.">
+<FaqItem id="invalid-or-missing-video-stream-in-segment" question="Invalid or missing video stream in segment. Discarding. / Discarding a corrupt recording segment / Failed to probe corrupt segment / Invalid recording segment detected">

-A cached recording segment failed validation (no readable video stream) and was deleted. The most common cause is a segment that was truncated because the record FFmpeg process was killed mid-write, so this often appears alongside, and as a consequence of, the record-stream restarts above. A segment containing only audio triggers it too.
+A cached recording segment failed validation and was deleted, either because it had no readable video stream or because its length was impossible. This nearly always means the camera stopped sending usable video partway through the segment: a camera that rebooted, dropped the connection, or ran out of simultaneous connections, or an unreliable link such as WiFi or a failing switch port. Broken camera timestamps (a "Smart Codec" / H.264+ mode) cause the corrupt-segment variants. The same stream failure trips the record watchdog, so the restarts above usually appear alongside these messages.
+
+See [Recordings: invalid or missing video stream in segment](/troubleshooting/recordings#invalid-or-missing-video-stream-in-segment).

 </FaqItem>

@@ -127,7 +133,7 @@ The process was killed by the CPU for executing an unsupported instruction. Ther

 <FaqItem id="onnx-invalidprotobuf" question="ONNX Runtime InvalidProtobuf / failed to load model">

-ONNX Runtime could not parse the model file. The file exists but its contents are not a valid ONNX model, usually a corrupted or interrupted download in `model_cache`, or the wrong file pointed at by `model.path`. Delete the cached model file so Frigate re-downloads it, and confirm `model.path` points at an actual `.onnx` model. See [ONNX detector configuration](/configuration/object_detectors#onnx).
+ONNX Runtime could not parse the model file. The file exists but its contents are not a valid ONNX model, usually a corrupted or interrupted download in `model_cache`, or the wrong file pointed at by a model's `path`. Delete the cached model file so Frigate re-downloads it, and confirm the model's `path` points at an actual `.onnx` model. See [ONNX detector configuration](/configuration/object_detectors#onnx).

 </FaqItem>

@@ -3,7 +3,31 @@ id: cpu
 title: High CPU Usage
 ---

-High CPU usage can impact Frigate's performance and responsiveness. This guide outlines the most effective configuration changes to help reduce CPU consumption and optimize resource usage.
+High CPU usage can impact Frigate's performance and responsiveness. This guide explains how to interpret the CPU values Frigate reports and outlines the most effective configuration changes to help reduce CPU consumption and optimize resource usage.
+
+## Understanding Frigate's Reported CPU Usage
+
+Frigate's CPU percentages often look much higher than what the host reports. Usually both numbers are correct and are simply measured against different denominators, so confirm you actually have a problem before tuning anything.
+
+### Per-process values are relative to a single core
+
+The values Frigate reports for FFmpeg, capture, detect, detector, and other processes follow the same convention as `top`: 100% means one CPU core is fully saturated, not that the whole system is saturated. A multithreaded process such as FFmpeg can legitimately report well over 100%.
+
+Host and hypervisor tools instead report a percentage of the machine's total capacity across all cores. This includes `docker stats`, the `htop` summary, the Proxmox summary graph, the Unraid dashboard, Synology Resource Monitor, and Home Assistant's system monitor sensors. To reconcile the two:
+
+```
+host percentage ≈ (sum of Frigate's process percentages) / (number of cores)
+```
+
+On a 4 core system, an FFmpeg process reporting 100% is consuming one quarter of the machine, so the host will show roughly 25 to 30% once the remaining Frigate processes are included. That same 100% on a 16 core system is about 6%. Frigate's own warning thresholds use the per-core convention as well, so an FFmpeg process is flagged at 20% of a single core, not 20% of the system.
+
+### Instantaneous samples and averages measure different things
+
+Frigate collects stats every 15 seconds, and the `cpu` value covers only the interval since the previous collection. The `cpu_average` value in the stats API and MQTT payload is the average across the entire life of the process, and it is what the high CPU usage warnings are based on. Host dashboards generally plot data averaged over a longer window, so a single Frigate sample can show a peak that a host graph never displays. A process that has just started, such as FFmpeg after a camera reconnect, reports 0 until it has been sampled twice.
+
+### The system-wide value depends on what the container can see
+
+The system CPU value is read from `/proc/stat`. Under Docker that file belongs to the host, so the value covers the entire machine including workloads unrelated to Frigate, and it will not match `docker stats` for the Frigate container. Under an LXC container, lxcfs virtualizes `/proc/stat` and the value reflects only the cores assigned to the container. In a virtual machine, the guest sees only its assigned vCPUs while the hypervisor divides by every physical thread on the node, so guest and host percentages will not agree even when both are accurate.

 ## 1. Hardware Acceleration for Video Decoding

@@ -72,3 +96,19 @@ The model you use significantly impacts detector performance. Frigate provides d
 - Larger models (640x640): Slower inference, can sometimes have higher accuracy on very large objects that take up a majority of the frame.

 For more detail on picking the right size, see [Choosing a model size](../configuration/object_detectors.md#choosing-a-model-size).
+
+## 3. Reducing Detector CPU Usage
+
+**Priority: High**
+
+The **Detector CPU Usage** metric measures the CPU spent converting frames into the tensor format the model expects and post-processing the model's output. It does not include inference, so this value can be high even when you've configured a GPU, NPU, or Coral for object detection.
+
+This metric scales with how many detections per second Frigate runs and how expensive each one is to prepare. Tuning [motion detection](../configuration/motion_detection) is usually the first recommendation to reduce the number of detections. Additionally, you can:
+
+- **Lower `detect -> fps`.** 5 is the recommended value for nearly all cameras. Running at 10 doubles the frames eligible for detection and is one of the largest contributors to this metric.
+- **Use a 320x320 model.** A 640x640 model has 4 times as many pixels to transpose, convert, and copy on every inference.
+- **Prefer a model that takes integer input.** Models configured with `input_dtype: float` require each frame to be converted to float32 and normalized on the CPU first. Models taking `int` input, such as the tflite models used by the Edge TPU, skip that step.
+- **Do not match the detect resolution to the model resolution.** The detect stream should match your camera's aspect ratio, for example `1280x720`, not the model's input size. Frigate crops and scales regions of motion itself, so an oversized detect stream only adds work.
+- **Tune stationary object behavior.** Objects that never settle into a stationary state are re-detected continuously. Raising `detect -> stationary -> interval` reduces how often detection runs on objects that are already parked. See [stationary objects](../configuration/stationary_objects).
+
+Adding [more detector instances](#multiple-detector-instances) spreads this work across more CPU cores, but does not reduce the total CPU used.
@@ -39,7 +39,7 @@ The per-clip variation is typically quite low and is mostly an artifact of keyfr

 Debug Replay lets you re-run Frigate's detection pipeline against a section of recorded video without manually configuring a dummy camera. It automatically extracts the recording, creates a temporary camera with the same detection settings as the original, and loops the clip through the pipeline so you can observe detections in real time.

-The replay camera behaves like a live camera feed rather than History's video player: it loops the clip continuously as Frigate analyzes it and has no playback controls, so you cannot pause, scrub, or step through it frame by frame.
+The replay camera behaves like a live camera feed rather than History's video player: it loops the clip continuously as Frigate analyzes it and has no playback controls, so you cannot pause, scrub, or step through it frame by frame. The Debug Replay camera does not save recordings or snapshots or surface anything in Explore, but it otherwise behaves like a regular camera, including running enrichments such as Face Recognition, LPR, and custom classification.

 Debug Replay isn't intended to be a one-stop pane for all Frigate diagnostics or a comprehensive debugging environment for every Frigate feature. It merely makes it easier to spin up a "dummy camera" and perform some common adjustments in real time. You'll still need to use the normal tools (logs, an MQTT client, etc) to debug your feature.

@@ -39,6 +39,20 @@ To do this efficiently the following setup is required:

 When this is done correctly, the GPU will do the decoding and scaling which will result in a small increase in CPU usage but with better results.

+### How can I rotate my camera's video feed?
+
+Rotation is best done in the camera's firmware settings (usually called rotate, flip, or corridor mode) so the video arrives already rotated and no extra processing is needed. Check there first.
+
+If your camera does not support rotation, go2rtc's ffmpeg module can rotate the stream with the `#rotate` parameter (`90`, `180`, `270`, or `-90`), but this is not recommended: rotation requires transcoding (re-encoding) the video, which significantly increases CPU usage, especially for high resolution streams.
+
+```yaml
+go2rtc:
+  streams:
+    my_camera: "ffmpeg:rtsp://user:password@192.168.1.10:554/stream#video=h264#hardware#rotate=90"
+```
+
+Point the camera's inputs at the restream as described in the [restream docs](/configuration/restream.md), and swap `detect -> width` and `detect -> height` to match the rotated resolution.
+
 ### My mjpeg stream or snapshots look green and crazy

 This almost always means that the width/height defined for your camera are not correct. Double check the resolution with VLC or another player. Also make sure you don't have the width and height values backwards.
@@ -65,9 +79,17 @@ This is because Frigate does not run in host mode so localhost points to the Fri

 ### How do I know if my camera is offline

-A camera being offline can be detected via MQTT or /api/stats, the camera_fps for any offline camera will be 0.
+Frigate publishes a per-role health status to [`frigate/<camera_name>/status/<role>`](/integrations/mqtt#frigatecamera_namestatusrole), where `<role>` is each enabled role on the camera (`detect`, `record`, and `audio`). The published value is one of:

-Also, Home Assistant will mark any offline camera as being unavailable when the camera is offline.
+- `online`: Frigate's process for that role is running normally
+- `offline`: the process is down and Frigate is restarting it
+- `disabled`: the camera is turned off, either at runtime or in the configuration file
+
+These reflect the state of Frigate's process for that role, not the camera's reachability, so an unreachable camera alternates between `offline` and `online` as the watchdog restarts ffmpeg. Wait for the status to hold steady (for example with Home Assistant's `for:`) rather than acting on a single message.
+
+Because the status is per role, a camera whose substream is fine but whose recording stream has dropped will report `online` for `detect` and `offline` for `record`. The status is republished whenever it changes.
+
+You can also detect an offline camera through `/api/stats`, where `camera_fps` will be 0.

 ### How can I view the Frigate log files without using the Web UI?

@@ -113,7 +135,7 @@ You can still configure Frigate to use UDP by using ffmpeg input args or the pre

 ### Frigate is slow to start up with a "probing detect stream" message in the logs

-When `detect.width` and `detect.height` are not set, Frigate probes each camera's detect stream on startup (and when saving the config) to auto-detect its resolution. For RTSP streams Frigate probes with ffprobe and automatically retries over TCP if UDP doesn't respond, with a 5 second timeout per attempt. A camera that cannot be reached over either transport will add up to ~10 seconds to startup before Frigate falls through with default dimensions, which may show up as width `0` and height `0` in Camera Probe Info under System Metrics.
+When `detect.width` and `detect.height` are not set, Frigate probes each camera's detect stream on startup (and when saving the config) to auto-detect its resolution. For RTSP streams Frigate probes with ffprobe and automatically retries over TCP if UDP doesn't respond, with a 5 second timeout per attempt. A camera that cannot be reached over either transport will add up to ~10 seconds to startup before Frigate falls through with default dimensions, which may show up as width `0` and height `0` in Camera Probe Info under Health and Metrics.

 To skip the probe entirely and make startup instant, set `detect.width` and `detect.height` explicitly in your camera config:

@@ -125,6 +147,12 @@ cameras:
      height: 720
 ```

+### What is the `version` key in my config file?
+
+`version` records the config format that your config was last migrated to. On startup Frigate compares it against the format the running version expects, and if it is older it copies your config to `/config/backup_config.yaml`, rewrites it to the new format, and updates `version` as the final step. A config with no `version` key is assumed to predate 0.14 and is migrated from there.
+
+Frigate manages this key for you, so do not set or edit it. Raising it makes Frigate skip migrations your config still needs, and lowering it re-runs migrations against config that has already been converted. Either can leave you with a config that no longer validates.
+
 ### Why does Frigate keep creating new tracked objects for my parked car?

 Stationary tracking is designed to _prevent_ this: a parked car should remain a single tracked object rather than generating new ones. If you're repeatedly getting new tracked objects for the same car, it's likely that Frigate is losing the object and re-detecting it as a new one.
@@ -153,3 +181,9 @@ Frigate's object detection relies on a machine learning [model](../frigate/gloss
 - If the false positive is always in the same fixed spot (like a statue or mailbox that reads as a person), add an [object filter mask](../configuration/masks.md#object-filter-masks) over that location.

 Filters and masks only hide the incorrect result - they don't teach Frigate what the object actually is. For that, fine-tune your own model or use Frigate+.
+
+### Where do I see problems Frigate has detected?
+
+Open System > Health. The Notices list keeps a record of problems Frigate has found, and you can dismiss any entry to acknowledge it. Ongoing conditions, such as an offline camera or recordings deleted before their retention period, appear in the status bar for admins until they clear, and the status bar links to the Notices list while it has undismissed entries. On mobile, tap the warning icon in the bottom navigation bar to see them.
+
+The Hardware section below the notices shows whether the detection hardware, hardware acceleration, and enrichment devices in your config were found and are being used, so a GPU that silently fell back to the CPU shows up as a warning. Run stream checks to probe every camera's streams for the same problems the camera wizard reports.
@@ -15,7 +15,7 @@ When a stream won't play or behaves oddly, the most important first step is to f

 ### 1. Read the go2rtc logs

-Access the go2rtc logs in the Frigate UI under <NavPath path="System Logs" /> in the sidebar (select the **go2rtc** tab). If go2rtc cannot connect to your camera you will usually see a clear error here: `401 Unauthorized` (bad or incorrectly encoded credentials), `Connection refused` / `timeout` (wrong IP, port, or the camera is at its connection limit), or `404 Not Found` (wrong RTSP path, or the referenced stream name does not exist).
+Access the go2rtc logs in the Frigate UI under <NavPath path="Logs" /> in the sidebar (select the **go2rtc** tab). If go2rtc cannot connect to your camera you will usually see a clear error here: `401 Unauthorized` (bad or incorrectly encoded credentials), `Connection refused` / `timeout` (wrong IP, port, or the camera is at its connection limit), or `404 Not Found` (wrong RTSP path, or the referenced stream name does not exist).

 ### 2. Test the stream in the go2rtc web interface

@@ -78,7 +78,9 @@ go2rtc:

 :::warning

-The `#`-modifiers (`#video=`, `#audio=`, `#hardware`, `#backchannel=0`, …) **only take effect on a source that is prefixed with `ffmpeg:`**. Adding them to a bare `rtsp://…#audio=opus` source does nothing: go2rtc ignores them. Likewise, when a source references another stream by name (e.g. `ffmpeg:back#audio=aac`), the name must match the stream key **exactly** (it is case sensitive), or the transcode is silently never produced. This is the single most common configuration mistake. In the Frigate UI, the **Use compatibility mode (ffmpeg)** toggle adds the `ffmpeg:` prefix for you.
+The transcoding modifiers (`#video=`, `#audio=`, `#hardware`, …) **only take effect on a source that is prefixed with `ffmpeg:`**. Adding them to a bare `rtsp://…#audio=opus` source does nothing: go2rtc ignores them. Likewise, when a source references another stream by name (e.g. `ffmpeg:back#audio=aac`), the name must match the stream key **exactly** (it is case sensitive), or the transcode is silently never produced. This is the single most common configuration mistake. In the Frigate UI, the **Use compatibility mode (ffmpeg)** toggle adds the `ffmpeg:` prefix for you.
+
+A bare `rtsp://` source reads a different set of modifiers: `#backchannel=`, `#media=`, `#timeout=`, and `#transport=`. These do nothing on an `ffmpeg:` source. Adding **any** modifier to a bare `rtsp://` source also disables the camera's backchannel unless the URL explicitly contains `#backchannel=1`, so a stream dedicated to two-way talk should carry no modifiers at all.

 :::

@@ -153,7 +155,7 @@ WebRTC is only attempted when MSE fails or when using a camera's two-way talk fe

 - **Codec mismatch**: WebRTC cannot carry H.265 or AAC. The stream backing the WebRTC view must provide Opus (or PCMA/PCMU) audio and H.264 video. Add an `ffmpeg:back#audio=opus` source as shown above.
 - **Port `8555` not reachable, or no candidates set**: WebRTC needs port `8555` (both TCP and UDP) open and a reachable candidate advertised. On Docker installs running on a custom/overlay network, go2rtc may advertise unreachable container IPs as ICE candidates; setting `webrtc.filters.candidates: []` and supplying only your host's LAN IP resolves this. See [WebRTC extra configuration](/configuration/live#webrtc-extra-configuration).
- **Two-way talk** additionally requires a secure context (HTTPS or the authenticated port `8971`, because browsers block microphone access on plain HTTP). The camera's RTSP backchannel must also be handled correctly: go2rtc seizes the backchannel by default, which blocks two-way audio for other consumers and can inject static. Disable it on the primary stream with `#backchannel=0` and use a separate dedicated stream for talk, as documented in [preventing go2rtc from blocking two-way audio](/configuration/restream#two-way-talk-restream).
+- **Two-way talk** additionally requires a secure context (HTTPS or the authenticated port `8971`, because browsers block microphone access on plain HTTP). The camera's RTSP backchannel must also be handled correctly: go2rtc seizes the backchannel by default, which blocks two-way audio for other consumers and can inject static. Disable it on the primary stream with `#backchannel=0` and use a separate dedicated stream for talk, carrying no `#` modifiers of any kind, as documented in [preventing go2rtc from blocking two-way audio](/configuration/restream#two-way-talk-restream).

 ## High CPU usage

@@ -209,6 +209,50 @@ If the record stream uses a "Smart Codec"/H.264+ mode or changes encoding parame

 </FaqItem>

+<FaqItem id="invalid-or-missing-video-stream-in-segment" question="I see the message: WARNING : Invalid or missing video stream in segment ... Discarding.">
+
+Every recording segment is validated before it leaves the cache. Frigate probes each finished `.mp4` in `/tmp/cache` and requires a readable video stream and a valid duration before moving to storage. A segment that fails is deleted, so those ~10 seconds of footage are lost. Three messages come from this check:
+
+- `Invalid or missing video stream in segment <path>. Discarding.` The segment holds no video, or could not be read at all.
+- `Failed to probe corrupt segment <path>` followed by `Discarding a corrupt recording segment: <path>`. The segment was read, but its length could not be determined.
+- `Discarding a corrupt recording segment: <path>` on its own. The segment's length is impossible (empty, or longer than ten minutes), which points at broken timestamps coming from the camera.
+
+For each one, the camera watchdog also logs `Invalid recording segment detected for <camera> at <timestamp>`.
+
+:::warning
+
+This is almost always a **camera or network problem**, not a Frigate one. A segment is only complete once ffmpeg has finished writing it, so anything that interrupts the stream partway through leaves behind a file that cannot be saved. Frigate is reporting the interruption, not causing it.
+
+:::
+
+#### Start with the camera and the network
+
+- **The camera dropped the connection.** Cameras reboot, reinitialize their stream when switching to night mode, and cut clients off when they are overloaded or out of simultaneous connections. Count everything pulling from the camera at once: Frigate's detect and record streams, go2rtc, a phone app, and any other NVR each use one. Routing all roles through a single [RTSP restream](/configuration/restream#reduce-connections-to-camera) so the camera only ever sees one connection often resolves this by itself.
+- **The link to the camera is unreliable.** WiFi cameras, powerline adapters, a saturated uplink, a failing switch port, or a marginal cable all produce this pattern, and usually only on one camera at a time. WiFi cameras are [not recommended](https://ipcamtalk.com/threads/multiple-cameras-high-bandwidth.77100/#post-861110).
+- **The camera cannot reliably send what it is being asked for.** A high bitrate 4K stream can be more than the camera's own hardware can encode and push out under load. Lower the bitrate, or record a lower-resolution profile.
+- **The camera is using a "Smart Codec", H.264+, or H.265+ mode.** These change encoding parameters mid-stream and produce the broken timestamps behind the corrupt-segment variant. Turn the mode off and set the camera's keyframe interval equal to its frame rate. See [Segments are only ~1 second long](#segments-are-only-1-second-long).
+
+Read the rest of the Frigate and/or go2rtc log around the **first** occurrence. When the camera or the network is at fault, other messages show up with it, such as `No frames received from <camera> in 20 seconds`, `Non-monotonic DTS`, `RTP: PT=xx: bad cseq`, `error while decoding MB`, or a connection timeout. Each of those is explained in [Common error messages](/troubleshooting/common_errors). To confirm the camera is the source, open its stream in the [go2rtc web interface](/troubleshooting/go2rtc) on port `1984` or play the same URL in VLC, and leave it running long enough for the failures to happen again.
+
+#### If the camera and network check out
+
+- **Audio the recording cannot store.** Some cameras send G.711 audio, which cannot be saved in an MP4 and stops segments from finalizing. See [Incompatible audio codec](#incompatible-audio-codec-recordings-silently-fail-to-save).
+- **Frigate itself was stopped or restarted.** A single warning per camera around a restart is expected and needs no action.
+- **The system ran out of room or memory.** A full `/tmp/cache`, or the host killing Frigate for using too much memory, cuts off the segment being written. Both leave other errors in the log alongside this one. See [No space left on device](#errno-28-no-space-left-on-device).
+
+</FaqItem>
+
+<FaqItem id="no-new-recording-segments-were-created" question="I see the message: ERROR : No new recording segments were created for <camera> in the last 120s. Restarting the ffmpeg record process...">
+
+When a camera stops producing usable recordings for two minutes, Frigate restarts that camera's record process to try to recover. The wording tells you how far the recordings got:
+
+- **`No new recording segments were created`**: no new segment file showed up in the cache at all, so ffmpeg isn't getting video out of the record stream. The camera is unreachable or refusing the connection, the stream URL, path, or credentials are wrong, or the camera accepted the connection and then sent nothing. See [The record stream isn't connecting](#the-record-stream-isnt-connecting).
+- **`No new valid recording segments were created`** and **`No valid segments created since last invalid segment`**: recordings are arriving, but they keep failing validation, so the camera is sending video that cannot be saved. See [Invalid or missing video stream in segment](#invalid-or-missing-video-stream-in-segment) above.
+
+The restart is Frigate recovering from a problem, not causing one. One of these after a camera reboot or a brief network drop is normal. Seeing them repeat every couple of minutes means the camera or the network is still failing, and the restarts can extend the damage, because each one cuts off the segment that was being written. Work from the earliest failure in that camera's log rather than from the restarts.
+
+</FaqItem>
+
 <FaqItem id="i-see-the-message-warning--unable-to-keep-up-with-recording-segments-in-cache-for-camera-keeping-the-5-most-recent-segments-out-of-6-and-discarding-the-rest" question="I see the message: WARNING : Unable to keep up with recording segments in cache for camera. Keeping the 5 most recent segments out of 6 and discarding the rest...">

 This warning means the recording maintainer cannot move recording segments from the RAM cache to disk fast enough. When the cache fills up, Frigate discards the oldest segments to avoid running out of memory and crashing, so you lose recorded footage. This is almost always a storage throughput or system resource problem. Work through the steps below to identify which.
@@ -330,7 +374,7 @@ If segments are only ~1 second instead of ~10 seconds, the camera is sending cor

 :::tip

-You don't have to run `ffprobe` by hand to catch this. Open a camera's **Camera Probe Info** dialog (the info icon on the System → Metrics → Cameras page) and check the **Keyframe analysis** section. It probes the record stream and flags sparse or variable keyframes, which is what smart/"+" codecs (H.264+/H.265+) and long keyframe intervals produce.
+You don't have to run `ffprobe` by hand to catch this. Open a camera's **Camera Probe Info** dialog (the info icon on the Health and Metrics → Cameras page) and check the **Keyframe analysis** section. It probes the record stream and flags sparse or variable keyframes, which is what smart/"+" codecs (H.264+/H.265+) and long keyframe intervals produce.

 :::

@@ -40,7 +40,7 @@ Deleting a group also clears any custom layout you saved for it.

 ## Rearranging a camera group layout

-On desktop and tablet, each camera group has its own freely-arrangeable grid. Enter **Edit Layout** mode from the layout button in the lower-right corner: camera tiles gain a drag handle and corner resize handles. Drag a tile to reposition it and drag a corner to resize it (the aspect ratio is preserved). Exit edit mode to save. The layout is stored in your browser per device, so each device can have its own arrangement.
+On desktop and tablet, each camera group has its own freely-arrangeable grid. Enter **Edit Layout** mode from the layout button in the lower-right corner: camera tiles gain a drag handle and corner resize handles. Drag a tile to reposition it and drag a corner to resize it (the aspect ratio is preserved). Exit edit mode to save. The layout is stored in your browser per device, so each device can have its own arrangement, and layouts can be exported to a file and imported on another device.

 The default **All Cameras** dashboard is not manually arrangeable. It automatically sizes tiles based on each camera's aspect ratio (wide cameras span two columns, tall cameras span two rows).

@@ -68,7 +68,7 @@ For non-default groups, the context menu also exposes **Streaming Settings** for
 - the **streaming method**: **No Streaming**, **Smart Streaming** (recommended), or **Continuous Streaming** (higher bandwidth), and
 - **compatibility mode**, for devices that have trouble rendering the default player.

-These settings are saved per group and per device in your browser, not in your config file.
+These settings are saved per group and per device in your browser, not in your config file, and can be exported to a file and imported on another device.

 ## The single-camera view

--- a/Show More
+++ b/Show More