Merge branch 'blakeblackshear:dev' into dev

.github/pull_request_template.md

@@ -26,7 +26,7 @@ _Please read the [contributing guidelines](https://github.com/blakeblackshear/fr
 
 - This PR fixes or closes issue: fixes #
 - This PR is related to issue:
-- Link to discussion with maintainers (**required** for large/pinned features):
+- Link to discussion with maintainers (**required** for any large or "planned" features):
 
 ## For new features
 

.github/workflows/pr_template_check.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check PR description against template
-        uses: actions/github-script@v7
+        uses: actions/github-script@v9
         with:
           script: |
             const maintainers = ['blakeblackshear', 'NickM-27', 'hawkeye217', 'dependabot[bot]', 'weblate'];

.github/workflows/pull_request.yml

@@ -72,7 +72,7 @@ jobs:
         run: npm run e2e
         working-directory: ./web
       - name: Upload test artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         if: failure()
         with:
           name: playwright-report

.github/workflows/stale.yml

@@ -18,9 +18,9 @@ jobs:
           close-issue-message: ""
           days-before-stale: 30
           days-before-close: 3
-          exempt-draft-pr: true
-          exempt-issue-labels: "pinned,security"
-          exempt-pr-labels: "pinned,security,dependencies"
+          exempt-draft-pr: false
+          exempt-issue-labels: "planned,security"
+          exempt-pr-labels: "planned,security,dependencies"
           operations-per-run: 120
       - name: Print outputs
         env:

.gitignore

@@ -22,3 +22,8 @@ core
 !/web/**/*.ts
 .idea/*
 .ipynb_checkpoints
+
+# Auto-generated Docker Compose Generator config files
+docs/src/components/DockerComposeGenerator/config/devices.ts
+docs/src/components/DockerComposeGenerator/config/hardware.ts
+docs/src/components/DockerComposeGenerator/config/ports.ts

CONTRIBUTING.md

@@ -10,11 +10,14 @@ If you've found a bug and want to fix it, go for it. Link to the relevant issue
 
 ### New features
 
-Every new feature adds scope that the maintainers must test, maintain, and support long-term. Before writing code for a new feature:
+A pull request is more than just code — it's a request for the maintainers to review, integrate, and support the change long-term. We're selective about what we take on, and prioritize changes that align with the project's direction and can be responsibly maintained in the long term.
 
-1. **Check for existing discussion.** Search [feature requests](https://github.com/blakeblackshear/frigate/issues) and [discussions](https://github.com/blakeblackshear/frigate/discussions) to see if it's been proposed or discussed. Pinned feature requests are on our radar — we plan to get to them, but we don't maintain a public roadmap or timeline. Check in with us first if you have interest in contributing to one.
+**Large or highly-requested features** raise the bar even higher. Popularity signals demand, but it doesn't pre-approve any particular implementation. The bigger the change, the higher the long-term cost, and the more important it is that we're aligned on scope and approach before any code is written. A large PR that lands without prior discussion is unlikely to be merged as-is, no matter how well it's implemented.
+
+Before writing code for a new feature:
+
+1. **Check for existing discussion.** Search [feature requests](https://github.com/blakeblackshear/frigate/issues) and [discussions](https://github.com/blakeblackshear/frigate/discussions) to see if it's been proposed or discussed. Feature requests tagged with "planned" are on our radar — we plan to get to them, but we don't maintain a public roadmap or timeline. Check in with us first if you have interest in contributing to one.
 2. **Start a discussion or feature request first.** This helps ensure your idea aligns with Frigate's direction before you invest time building it. Community interest in a feature request helps us gauge demand, though a great idea is a great idea even without a crowd behind it.
-3. **Be open to "no".** We try to be thoughtful about what we take on, and sometimes that means saying no to good code if the feature isn't the right fit for the project. These calls are sometimes subjective, and we won't always get them right. We're happy to discuss and reconsider.
 
 ## AI usage policy
 
@@ -39,6 +42,8 @@ We're not trying to gatekeep how you write code. Use whatever tools make you pro
 
 Some honest context: when we review a PR, we're not just evaluating whether the code works today. We're evaluating whether we can maintain it, debug it, and extend it long-term — often without the original author's involvement. Code that the author doesn't deeply understand is code that nobody understands, and that's a liability.
 
+One more thing worth saying directly: most maintainers already have access to the same AI tools you do. A PR that's entirely AI-generated — where the author can't explain the design, debug issues independently, or engage substantively in design discussions — doesn't offer something we couldn't produce ourselves. What makes a contribution genuinely valuable is the human judgment and domain understanding behind it, as well as the engagement during review that shapes it into something we can confidently take on long-term.
+
 ## Pull request guidelines
 
 ### Before submitting

docker/main/install_deps.sh

@@ -87,43 +87,43 @@ if [[ "${TARGETARCH}" == "amd64" ]]; then
     # intel packages use zst compression so we need to update dpkg
     apt-get install -y dpkg
 
-    # use intel apt intel packages
+    # use intel apt repo for libmfx1 (legacy QSV, pre-Gen12)
     wget -qO - https://repositories.intel.com/gpu/intel-graphics.key | gpg --yes --dearmor --output /usr/share/keyrings/intel-graphics.gpg
     echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu jammy client" | tee /etc/apt/sources.list.d/intel-gpu-jammy.list
     apt-get -qq update
+
     # intel-media-va-driver-non-free is built from source in the
     # intel-media-driver Dockerfile stage for Battlemage (Xe2) support
     apt-get -qq install --no-install-recommends --no-install-suggests -y \
-        libmfx1 libmfxgen1 libvpl2
+        libmfx1
+    rm -f /usr/share/keyrings/intel-graphics.gpg
+    rm -f /etc/apt/sources.list.d/intel-gpu-jammy.list
 
+    # upgrade libva2, oneVPL runtime, and libvpl2 from trixie for Battlemage support
+    echo "deb http://deb.debian.org/debian trixie main" > /etc/apt/sources.list.d/trixie.list
+    apt-get -qq update
+    apt-get -qq install -y -t trixie libva2 libva-drm2 libzstd1
+    apt-get -qq install -y -t trixie libmfx-gen1.2 libvpl2
+    rm -f /etc/apt/sources.list.d/trixie.list
+    apt-get -qq update
     apt-get -qq install -y ocl-icd-libopencl1
 
     # install libtbb12 for NPU support
     apt-get -qq install -y libtbb12
 
-    rm -f /usr/share/keyrings/intel-graphics.gpg
-    rm -f /etc/apt/sources.list.d/intel-gpu-jammy.list
-
-    # install legacy and standard intel icd and level-zero-gpu
+    # install legacy and standard intel compute packages
     # see https://github.com/intel/compute-runtime/blob/master/LEGACY_PLATFORMS.md for more info
-    # newer intel packages (gmmlib 22.9+, igc 2.32+) require libstdc++ >= 13.1 and libzstd >= 1.5.5
-    echo "deb http://deb.debian.org/debian trixie main" > /etc/apt/sources.list.d/trixie.list
-    apt-get -qq update
-    apt-get -qq install -y -t trixie libstdc++6 libzstd1
-    rm -f /etc/apt/sources.list.d/trixie.list
-    apt-get -qq update
-
     # needed core package
     wget 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 packages
+    # 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
-    # standard packages
+    # 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
@@ -137,6 +137,10 @@ if [[ "${TARGETARCH}" == "amd64" ]]; then
     dpkg -i *.deb
     rm *.deb
     apt-get -qq install -f -y
+
+    # Battlemage uses the xe kernel driver, but the VA-API driver is still iHD.
+    # The oneVPL runtime may look for a driver named after the kernel module.
+    ln -sf /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so /usr/lib/x86_64-linux-gnu/dri/xe_drv_video.so
 fi
 
 if [[ "${TARGETARCH}" == "arm64" ]]; then

docker/main/requirements-wheels.txt

@@ -11,7 +11,7 @@ joserfc == 1.2.*
 cryptography == 44.0.*
 pathvalidate == 3.3.*
 markupsafe == 3.0.*
-python-multipart == 0.0.20
+python-multipart == 0.0.26
 # Classification Model Training
 tensorflow == 2.19.* ; platform_machine == 'aarch64'
 tensorflow-cpu == 2.19.* ; platform_machine == 'x86_64'
@@ -42,7 +42,7 @@ opencv-python-headless == 4.11.0.*
 opencv-contrib-python == 4.11.0.*
 scipy == 1.16.*
 # OpenVino & ONNX
-openvino == 2025.3.*
+openvino == 2025.4.*
 onnxruntime == 1.22.*
 # Embeddings
 transformers == 4.45.*

docker/main/rootfs/usr/local/go2rtc/create_config.py

@@ -3,7 +3,6 @@
 import json
 import os
 import sys
-from pathlib import Path
 from typing import Any
 
 from ruamel.yaml import YAML
@@ -18,37 +17,12 @@ from frigate.const import (
 )
 from frigate.ffmpeg_presets import parse_preset_hardware_acceleration_encode
 from frigate.util.config import find_config_file
+from frigate.util.services import is_restricted_go2rtc_source
 
 sys.path.remove("/opt/frigate")
 
 yaml = YAML()
 
-# Check if arbitrary exec sources are allowed (defaults to False for security)
-allow_arbitrary_exec = None
-if "GO2RTC_ALLOW_ARBITRARY_EXEC" in os.environ:
-    allow_arbitrary_exec = os.environ.get("GO2RTC_ALLOW_ARBITRARY_EXEC")
-elif (
-    os.path.isdir("/run/secrets")
-    and os.access("/run/secrets", os.R_OK)
-    and "GO2RTC_ALLOW_ARBITRARY_EXEC" in os.listdir("/run/secrets")
-):
-    allow_arbitrary_exec = (
-        Path(os.path.join("/run/secrets", "GO2RTC_ALLOW_ARBITRARY_EXEC"))
-        .read_text()
-        .strip()
-    )
-# check for the add-on options file
-elif os.path.isfile("/data/options.json"):
-    with open("/data/options.json") as f:
-        raw_options = f.read()
-    options = json.loads(raw_options)
-    allow_arbitrary_exec = options.get("go2rtc_allow_arbitrary_exec")
-
-ALLOW_ARBITRARY_EXEC = allow_arbitrary_exec is not None and str(
-    allow_arbitrary_exec
-).lower() in ("true", "1", "yes")
-
-
 config_file = find_config_file()
 
 try:
@@ -128,18 +102,13 @@ if LIBAVFORMAT_VERSION_MAJOR < 59:
         go2rtc_config["ffmpeg"]["rtsp"] = rtsp_args
 
 
-def is_restricted_source(stream_source: str) -> bool:
-    """Check if a stream source is restricted (echo, expr, or exec)."""
-    return stream_source.strip().startswith(("echo:", "expr:", "exec:"))
-
-
 for name in list(go2rtc_config.get("streams", {})):
     stream = go2rtc_config["streams"][name]
 
     if isinstance(stream, str):
         try:
             formatted_stream = substitute_frigate_vars(stream)
-            if not ALLOW_ARBITRARY_EXEC and is_restricted_source(formatted_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. "
                     f"Set GO2RTC_ALLOW_ARBITRARY_EXEC=true to enable arbitrary exec sources."
@@ -158,7 +127,7 @@ for name in list(go2rtc_config.get("streams", {})):
         for i, stream_item in enumerate(stream):
             try:
                 formatted_stream = substitute_frigate_vars(stream_item)
-                if not ALLOW_ARBITRARY_EXEC and is_restricted_source(formatted_stream):
+                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. "
                         f"Set GO2RTC_ALLOW_ARBITRARY_EXEC=true to enable arbitrary exec sources."

docker/rocm/Dockerfile

@@ -13,7 +13,7 @@ ARG ROCM
 
 RUN apt update -qq && \
     apt install -y wget gpg && \
-    wget -O rocm.deb https://repo.radeon.com/amdgpu-install/7.2/ubuntu/jammy/amdgpu-install_7.2.70200-1_all.deb && \
+    wget -O rocm.deb https://repo.radeon.com/amdgpu-install/7.2.3/ubuntu/jammy/amdgpu-install_7.2.3.70203-1_all.deb && \
     apt install -y ./rocm.deb && \
     apt update && \
     apt install -qq -y rocm
@@ -32,11 +32,14 @@ RUN echo /opt/rocm/lib|tee /opt/rocm-dist/etc/ld.so.conf.d/rocm.conf
 FROM deps AS deps-prelim
 
 COPY docker/rocm/debian-backports.sources /etc/apt/sources.list.d/debian-backports.sources
-RUN apt-get update && \
+# install_deps.sh upgraded libstdc++6 from trixie for Battlemage; the matching
+# -dev package must also come from trixie or apt refuses to satisfy it.
+RUN echo "deb http://deb.debian.org/debian trixie main" > /etc/apt/sources.list.d/trixie.list && \
+    apt-get update && \
     apt-get install -y libnuma1 && \
     apt-get install -qq -y -t bookworm-backports mesa-va-drivers mesa-vulkan-drivers && \
-    # Install C++ standard library headers for HIPRTC kernel compilation fallback
-    apt-get install -qq -y libstdc++-12-dev && \
+    apt-get install -qq -y -t trixie libstdc++-14-dev && \
+    rm -f /etc/apt/sources.list.d/trixie.list && \
     rm -rf /var/lib/apt/lists/*
 
 WORKDIR /opt/frigate
@@ -75,6 +78,10 @@ ENV MIGRAPHX_DISABLE_MIOPEN_FUSION=1
 ENV MIGRAPHX_DISABLE_SCHEDULE_PASS=1
 ENV MIGRAPHX_DISABLE_REDUCE_FUSION=1
 ENV MIGRAPHX_ENABLE_HIPRTC_WORKAROUNDS=1
+ENV MIOPEN_CUSTOM_CACHE_DIR=/config/model_cache/migraphx
+ENV MIOPEN_USER_DB_PATH=/config/model_cache/migraphx
+ENV AMD_COMGR_CACHE=1
+ENV AMD_COMGR_CACHE_DIR=/config/model_cache/migraphx
 
 COPY --from=rocm-dist / /
 

docker/rocm/requirements-wheels-rocm.txt

@@ -1 +1 @@
-onnxruntime-migraphx @ https://github.com/NickM-27/frigate-onnxruntime-rocm/releases/download/v7.2.0/onnxruntime_migraphx-1.23.1-cp311-cp311-linux_x86_64.whl
+onnxruntime-migraphx @ https://github.com/NickM-27/frigate-onnxruntime-rocm/releases/download/v7.2.3-1/onnxruntime_migraphx-1.24.4-cp311-cp311-linux_x86_64.whl

docker/rocm/rocm.hcl

@@ -1,5 +1,5 @@
 variable "ROCM" {
-  default = "7.2.0"
+  default = "7.2.3"
 }
 variable "HSA_OVERRIDE_GFX_VERSION" {
   default = ""

docs/docs/configuration/face_recognition.md

@@ -19,7 +19,7 @@ Face recognition requires a one-time internet connection to download detection a
 
 ### Face Detection
 
-When running a Frigate+ model (or any custom model that natively detects faces) should ensure that `face` is added to the [list of objects to track](../plus/#available-label-types) either globally or for a specific camera. This will allow face detection to run at the same time as object detection and be more efficient.
+When running a Frigate+ model (or any custom model that natively detects faces) should ensure that `face` is added to the [list of objects to track](../plus/index.md#available-label-types) either globally or for a specific camera. This will allow face detection to run at the same time as object detection and be more efficient.
 
 When running a default COCO model or another model that does not include `face` as a detectable label, face detection will run via CV2 using a lightweight DNN model that runs on the CPU. In this case, you should _not_ define `face` in your list of objects to track.
 
@@ -171,7 +171,7 @@ When choosing images to include in the face training set it is recommended to al
 - If it is difficult to make out details in a persons face it will not be helpful in training.
 - Avoid images with extreme under/over-exposure.
 - Avoid blurry / pixelated images.
-- Avoid training on infrared (gray-scale). The models are trained on color images and will be able to extract features from gray-scale images.
+- Avoid training on infrared (gray-scale). The models are trained on color images and will not be able to extract features from gray-scale images.
 - Using images of people wearing hats / sunglasses may confuse the model.
 - Do not upload too many similar images at the same time, it is recommended to train no more than 4-6 similar images for each person to avoid over-fitting.
 

docs/docs/configuration/genai/config.md

@@ -201,7 +201,7 @@ Cloud Generative AI providers require an active internet connection to send imag
 
 ### Ollama Cloud
 
-Ollama also supports [cloud models](https://ollama.com/cloud), where your local Ollama instance handles requests from Frigate, but model inference is performed in the cloud. Set up Ollama locally, sign in with your Ollama account, and specify the cloud model name in your Frigate config. For more details, see the Ollama cloud model [docs](https://docs.ollama.com/cloud).
+Ollama also supports [cloud models](https://ollama.com/cloud), where model inference is performed in the cloud. You can connect directly to Ollama Cloud by setting `base_url` to `https://ollama.com` and providing an API key. Alternatively, you can run Ollama locally and use a cloud model name so your local instance forwards requests to the cloud. For more details, see the Ollama cloud model [docs](https://docs.ollama.com/cloud).
 
 #### Configuration
 
@@ -210,7 +210,8 @@ Ollama also supports [cloud models](https://ollama.com/cloud), where your local
 
 1. Navigate to <NavPath path="Settings > Enrichments > Generative AI" />.
    - Set **Provider** to `ollama`
-   - Set **Base URL** to your local Ollama address (e.g., `http://localhost:11434`)
+   - Set **Base URL** to your local Ollama address (e.g., `http://localhost:11434`) or `https://ollama.com` for direct cloud inference
+   - Set **API key** if required by your endpoint (e.g., when using `https://ollama.com`)
    - Set **Model** to the cloud model name
 
 </TabItem>
@@ -223,6 +224,16 @@ genai:
   model: cloud-model-name
 ```
 
+or when using Ollama Cloud directly
+
+```yaml
+genai:
+  provider: ollama
+  base_url: https://ollama.com
+  model: cloud-model-name
+  api_key: your-api-key
+```
+
 </TabItem>
 </ConfigTabs>
 

docs/docs/configuration/hardware_acceleration_video.md

@@ -136,90 +136,32 @@ ffmpeg:
 </TabItem>
 </ConfigTabs>
 
-### Configuring Intel GPU Stats in Docker
+### Configuring Intel GPU Stats
 
-Additional configuration is needed for the Docker container to be able to access the `intel_gpu_top` command for GPU stats. There are two options:
+Frigate reads Intel GPU utilization directly from the kernel's per-client DRM usage counters exposed at `/proc/<pid>/fdinfo/<fd>`. This requires:
 
-1. Run the container as privileged.
-2. Add the `CAP_PERFMON` capability (note: you might need to set the `perf_event_paranoid` low enough to allow access to the performance event system.)
+- Linux kernel **5.19 or newer** for the `i915` driver, or any release of the `xe` driver.
+- Frigate running with permission to read other processes' fdinfo. Running as root inside the container (the default) satisfies this; non-root setups may need `CAP_SYS_PTRACE`.
 
-#### Run as privileged
+No `intel_gpu_top` binary, `CAP_PERFMON`, privileged mode, or `perf_event_paranoid` tuning is required.
 
-This method works, but it gives more permissions to the container than are actually needed.
+#### Stats for SR-IOV or specific devices
 
-##### Docker Compose - Privileged
-
-```yaml
-services:
-  frigate:
-    ...
-    image: ghcr.io/blakeblackshear/frigate:stable
-    # highlight-next-line
-    privileged: true
-```
-
-##### Docker Run CLI - Privileged
-
-```bash {4}
-docker run -d \
-  --name frigate \
-  ...
-  --privileged \
-  ghcr.io/blakeblackshear/frigate:stable
-```
-
-#### CAP_PERFMON
-
-Only recent versions of Docker support the `CAP_PERFMON` capability. You can test to see if yours supports it by running: `docker run --cap-add=CAP_PERFMON hello-world`
-
-##### Docker Compose - CAP_PERFMON
-
-```yaml {5,6}
-services:
-  frigate:
-    ...
-    image: ghcr.io/blakeblackshear/frigate:stable
-    cap_add:
-      - CAP_PERFMON
-```
-
-##### Docker Run CLI - CAP_PERFMON
-
-```bash {4}
-docker run -d \
-  --name frigate \
-  ...
-  --cap-add=CAP_PERFMON \
-  ghcr.io/blakeblackshear/frigate:stable
-```
-
-#### perf_event_paranoid
-
-_Note: This setting must be changed for the entire system._
-
-For more information on the various values across different distributions, see https://askubuntu.com/questions/1400874/what-does-perf-paranoia-level-four-do.
-
-Depending on your OS and kernel configuration, you may need to change the `/proc/sys/kernel/perf_event_paranoid` kernel tunable. You can test the change by running `sudo sh -c 'echo 2 >/proc/sys/kernel/perf_event_paranoid'` which will persist until a reboot. Make it permanent by running `sudo sh -c 'echo kernel.perf_event_paranoid=2 >> /etc/sysctl.d/local.conf'`
-
-#### Stats for SR-IOV or other devices
-
-When using virtualized GPUs via SR-IOV, you need to specify the device path to use to gather stats from `intel_gpu_top`. This example may work for some systems using SR-IOV:
+If the host has more than one Intel GPU (e.g. an iGPU plus a discrete GPU, or SR-IOV virtual functions), pin stats collection to a specific device by setting `intel_gpu_device` to either its PCI bus address or a DRM card/render-node path:
 
 ```yaml
 telemetry:
   stats:
-    intel_gpu_device: "sriov"
+    intel_gpu_device: "0000:00:02.0"
 ```
 
-For other virtualized GPUs, try specifying the direct path to the device instead:
-
 ```yaml
 telemetry:
   stats:
-    intel_gpu_device: "drm:/dev/dri/card0"
+    intel_gpu_device: "/dev/dri/card1"
 ```
 
-If you are passing in a device path, make sure you've passed the device through to the container.
+When passing a device path, make sure the device is also passed through to the container.
 
 ## AMD-based CPUs
 

docs/docs/configuration/object_detectors.md

@@ -72,7 +72,7 @@ This does not affect using hardware for accelerating other tasks such as [semant
 
 # 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 OpenVINO detector running on the CPU. 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.
 
 ## Edge TPU Detector
 
@@ -494,7 +494,7 @@ detectors:
 | [YOLO-NAS](#yolo-nas)                 | ✅  | ✅  |                                                              |
 | [MobileNet v2](#ssdlite-mobilenet-v2) | ✅  | ✅  | Fast and lightweight model, less accurate than larger models |
 | [YOLOX](#yolox)                       | ✅  | ?   |                                                              |
-| [D-FINE](#d-fine)                     | ❌  | ❌  |                                                              |
+| [D-FINE / DEIMv2](#d-fine--deimv2)    | ❌  | ❌  |                                                              |
 
 #### SSDLite MobileNet v2
 
@@ -710,13 +710,13 @@ model:
 
 </details>
 
-#### D-FINE
+#### D-FINE / DEIMv2
 
-[D-FINE](https://github.com/Peterande/D-FINE) is a DETR based model. The ONNX exported models are supported, but not included by default. See [the models section](#downloading-d-fine-model) for more information on downloading the D-FINE model for use in Frigate.
+[D-FINE](https://github.com/Peterande/D-FINE) and [DEIMv2](https://github.com/Intellindust-AI-Lab/DEIMv2) are DETR based models that share the same ONNX input/output format. The ONNX exported models are supported, but not included by default. See the models section for downloading [D-FINE](#downloading-d-fine-model) or [DEIMv2](#downloading-deimv2-model) for use in Frigate.
 
 :::warning
 
-Currently D-FINE models only run on OpenVINO in CPU mode, GPUs currently fail to compile the model
+Currently D-FINE / DEIMv2 models only run on OpenVINO in CPU mode, GPUs currently fail to compile the model
 
 :::
 
@@ -766,6 +766,31 @@ Note that the labelmap uses a subset of the complete COCO label set that has onl
 
 </details>
 
+<details>
+  <summary>DEIMv2 Setup & Config</summary>
+
+After placing the downloaded onnx model in your `config/model_cache` folder, you can use the following configuration:
+
+```yaml
+detectors:
+  ov:
+    type: openvino
+    device: CPU
+
+model:
+  model_type: dfine
+  width: 640
+  height: 640
+  input_tensor: nchw
+  input_dtype: float
+  path: /config/model_cache/deimv2_hgnetv2_n.onnx
+  labelmap_path: /labelmap/coco-80.txt
+```
+
+Note that the labelmap uses a subset of the complete COCO label set that has only 80 objects.
+
+</details>
+
 ## Apple Silicon detector
 
 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`.
@@ -947,7 +972,7 @@ The AMD GPU kernel is known problematic especially when converting models to mxr
 
 See [ONNX supported models](#supported-models) for supported models, there are some caveats:
 
-- D-FINE models are not supported
+- D-FINE / DEIMv2 models are not supported
 - YOLO-NAS models are known to not run well on integrated GPUs
 
 ## ONNX
@@ -997,13 +1022,13 @@ detectors:
 
 ### ONNX Supported Models
 
-| Model                         | Nvidia GPU | AMD GPU | Notes                                               |
-| ----------------------------- | ---------- | ------- | --------------------------------------------------- |
-| [YOLOv9](#yolo-v3-v4-v7-v9-2) | ✅         | ✅      | Supports CUDA Graphs for optimal Nvidia performance |
-| [RF-DETR](#rf-detr)           | ✅         | ❌      | Supports CUDA Graphs for optimal Nvidia performance |
-| [YOLO-NAS](#yolo-nas-1)       | ⚠️         | ⚠️      | Not supported by CUDA Graphs                        |
-| [YOLOX](#yolox-1)             | ✅         | ✅      | Supports CUDA Graphs for optimal Nvidia performance |
-| [D-FINE](#d-fine)             | ⚠️         | ❌      | Not supported by CUDA Graphs                        |
+| Model                                | Nvidia GPU | AMD GPU | Notes                                               |
+| ------------------------------------ | ---------- | ------- | --------------------------------------------------- |
+| [YOLOv9](#yolo-v3-v4-v7-v9-2)        | ✅         | ✅      | Supports CUDA Graphs for optimal Nvidia performance |
+| [RF-DETR](#rf-detr)                  | ✅         | ⚠️      | Supports CUDA Graphs for optimal Nvidia performance |
+| [YOLO-NAS](#yolo-nas-1)              | ⚠️         | ⚠️      | Not supported by CUDA Graphs                        |
+| [YOLOX](#yolox-1)                    | ✅         | ✅      | Supports CUDA Graphs for optimal Nvidia performance |
+| [D-FINE / DEIMv2](#d-fine--deimv2-1) | ⚠️         | ❌      | Not supported by CUDA Graphs                        |
 
 There is no default model provided, the following formats are supported:
 
@@ -1215,9 +1240,9 @@ model:
 
 </details>
 
-#### D-FINE
+#### D-FINE / DEIMv2
 
-[D-FINE](https://github.com/Peterande/D-FINE) is a DETR based model. The ONNX exported models are supported, but not included by default. See [the models section](#downloading-d-fine-model) for more information on downloading the D-FINE model for use in Frigate.
+[D-FINE](https://github.com/Peterande/D-FINE) and [DEIMv2](https://github.com/Intellindust-AI-Lab/DEIMv2) are DETR based models that share the same ONNX input/output format. The ONNX exported models are supported, but not included by default. See the models section for downloading [D-FINE](#downloading-d-fine-model) or [DEIMv2](#downloading-deimv2-model) for use in Frigate.
 
 <details>
   <summary>D-FINE Setup & Config</summary>
@@ -1262,6 +1287,28 @@ model:
 
 </details>
 
+<details>
+  <summary>DEIMv2 Setup & Config</summary>
+
+After placing the downloaded onnx model in your `config/model_cache` folder, you can use the following configuration:
+
+```yaml
+detectors:
+  onnx:
+    type: onnx
+
+model:
+  model_type: dfine
+  width: 640
+  height: 640
+  input_tensor: nchw
+  input_dtype: float
+  path: /config/model_cache/deimv2_hgnetv2_n.onnx
+  labelmap_path: /labelmap/coco-80.txt
+```
+
+</details>
+
 Note that the labelmap uses a subset of the complete COCO label set that has only 80 objects.
 
 ## CPU Detector (not recommended)
@@ -1405,7 +1452,7 @@ MemryX `.dfp` models are automatically downloaded at runtime, if enabled, to the
 
 #### YOLO-NAS
 
-The [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md) model included in this detector is downloaded from the [Models Section](#downloading-yolo-nas-model) and compiled to DFP with [mx_nc](https://developer.memryx.com/tools/neural_compiler.html#usage).
+The [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md) model included in this detector is downloaded from the [Models Section](#downloading-yolo-nas-model) and compiled to DFP with [mx_nc](https://developer.memryx.com/2p1/tools/neural_compiler.html#usage).
 
 **Note:** The default model for the MemryX detector is YOLO-NAS 320x320.
 
@@ -1459,7 +1506,7 @@ model:
 
 #### YOLOv9
 
-The YOLOv9s model included in this detector is downloaded from [the original GitHub](https://github.com/WongKinYiu/yolov9) like in the [Models Section](#yolov9-1) and compiled to DFP with [mx_nc](https://developer.memryx.com/tools/neural_compiler.html#usage).
+The YOLOv9s model included in this detector is downloaded from [the original GitHub](https://github.com/WongKinYiu/yolov9) like in the [Models Section](#yolov9-1) and compiled to DFP with [mx_nc](https://developer.memryx.com/2p1/tools/neural_compiler.html#usage).
 
 ##### Configuration
 
@@ -1601,19 +1648,39 @@ model:
 
 #### Using a Custom Model
 
-To use your own model:
+To use your own custom model, first compile it into a [.dfp](https://developer.memryx.com/2p1/specs/files.html#dataflow-program) file, which is the format used by MemryX.
 
-1.  Package your compiled model into a `.zip` file.
+#### Compile the Model
 
-2.  The `.zip` must contain the compiled `.dfp` file.
+Custom models must be compiled using **MemryX SDK 2.1**.
 
-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`.
+Before compiling your model, install the MemryX Neural Compiler tools from the
+[Install Tools](https://developer.memryx.com/2p1/get_started/install_tools.html) page on the **host**.
 
-4.  Bind-mount the `.zip` file into the container and specify its path using `model.path` in your config.
+> **Note:** It is recommended to compile the model on the host machine, or on another separate machine, rather than inside the Frigate Docker container. Installing the compiler inside Docker may conflict with container packages. It is recommended to create a Python virtual environment and install the compiler there.
 
-5.  Update the `labelmap_path` to match your custom model's labels.
+Once the SDK 2.1 environment is set up, follow the
+[MemryX Compiler](https://developer.memryx.com/2p1/tools/neural_compiler.html#usage) documentation to compile your model.
 
-For detailed instructions on compiling models, refer to the [MemryX Compiler](https://developer.memryx.com/tools/neural_compiler.html#usage) docs and [Tutorials](https://developer.memryx.com/tutorials/tutorials.html).
+Example:
+
+```bash
+mx_nc -m yolonas.onnx -c 4 --autocrop -v --dfp_fname yolonas.dfp
+```
+
+For detailed instructions on compiling models, refer to the [MemryX Compiler](https://developer.memryx.com/2p1/tools/neural_compiler.html#usage) docs and [Tutorials](https://developer.memryx.com/2p1/tutorials/tutorials.html).
+
+#### Package the Compiled Model
+
+1. Package your compiled model into a `.zip` file.
+
+2. The `.zip` file must contain the compiled `.dfp` file.
+
+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.
+
+5. Update `labelmap_path` to match your custom model's labels.
 
 ```yaml
 # The detector automatically selects the default model if nothing is provided in the config.
@@ -2274,6 +2341,49 @@ COPY --from=build /dfine/output/dfine_${MODEL_SIZE}_obj2coco.onnx /dfine-${MODEL
 EOF
 ```
 
+### Downloading DEIMv2 Model
+
+[DEIMv2](https://github.com/Intellindust-AI-Lab/DEIMv2) can be exported as ONNX by running the command below. Pretrained weights are available on Hugging Face for two backbone families:
+
+- **HGNetv2** (smaller/faster): `atto`, `femto`, `pico`, `n`
+- **DINOv3** (larger/more accurate): `s`, `m`, `l`, `x`
+
+Set `BACKBONE` and `MODEL_SIZE` in the first line to match your desired variant. Hugging Face model names use uppercase (e.g. `HGNetv2_N`, `DINOv3_S`), while config files use lowercase (e.g. `hgnetv2_n`, `dinov3_s`).
+
+```sh
+docker build . --rm --build-arg BACKBONE=hgnetv2 --build-arg MODEL_SIZE=n --output . -f- <<'EOF'
+FROM python:3.11-slim AS build
+RUN apt-get update && apt-get install --no-install-recommends -y git libgl1 libglib2.0-0 && rm -rf /var/lib/apt/lists/*
+COPY --from=ghcr.io/astral-sh/uv:0.8.0 /uv /bin/
+WORKDIR /deimv2
+RUN git clone https://github.com/Intellindust-AI-Lab/DEIMv2.git .
+# Install CPU-only PyTorch first to avoid pulling CUDA variant
+RUN uv pip install --no-cache --system torch torchvision --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache --system -r requirements.txt
+RUN uv pip install --no-cache --system onnx safetensors huggingface_hub
+RUN mkdir -p output
+ARG BACKBONE
+ARG MODEL_SIZE
+# Download from Hugging Face and convert safetensors to pth
+RUN python3 -c "\
+from huggingface_hub import hf_hub_download; \
+from safetensors.torch import load_file; \
+import torch; \
+backbone = '${BACKBONE}'.replace('hgnetv2','HGNetv2').replace('dinov3','DINOv3'); \
+size = '${MODEL_SIZE}'.upper(); \
+st = load_file(hf_hub_download('Intellindust/DEIMv2_' + backbone + '_' + size + '_COCO', 'model.safetensors')); \
+torch.save({'model': st}, 'output/deimv2.pth')"
+RUN sed -i "s/data = torch.rand(2/data = torch.rand(1/" tools/deployment/export_onnx.py
+# HuggingFace safetensors omits frozen constants that the model constructor initializes
+RUN sed -i "s/cfg.model.load_state_dict(state)/cfg.model.load_state_dict(state, strict=False)/" tools/deployment/export_onnx.py
+RUN python3 tools/deployment/export_onnx.py -c configs/deimv2/deimv2_${BACKBONE}_${MODEL_SIZE}_coco.yml -r output/deimv2.pth
+FROM scratch
+ARG BACKBONE
+ARG MODEL_SIZE
+COPY --from=build /deimv2/output/deimv2.onnx /deimv2_${BACKBONE}_${MODEL_SIZE}.onnx
+EOF
+```
+
 ### Downloading RF-DETR Model
 
 RF-DETR can be exported as ONNX by running the command below. You can copy and paste the whole thing to your terminal and execute, altering `MODEL_SIZE=Nano` in the first line to `Nano`, `Small`, or `Medium` size.

docs/docs/configuration/record.md

@@ -195,7 +195,7 @@ Pre and post capture footage is included in the **recording timeline**, visible
 
 ## Will Frigate delete old recordings if my storage runs out?
 
-As of Frigate 0.12 if there is less than an hour left of storage, the oldest 2 hours of recordings will be deleted.
+If there is less than an hour left of storage, the oldest hour of recordings will be deleted and a message will be printed in the Frigate logs. This emergency cleanup deletes the oldest recordings first regardless of retention settings to reclaim space as quickly as possible.
 
 ## Configuring Recording Retention
 

docs/docs/configuration/restream.md

@@ -236,7 +236,7 @@ Enabling arbitrary exec sources allows execution of arbitrary commands through g
 
 ## Advanced Restream Configurations
 
-The [exec](https://github.com/AlexxIT/go2rtc/tree/v1.9.13#source-exec) source in go2rtc can be used for custom ffmpeg commands. An example is below:
+The [exec](https://github.com/AlexxIT/go2rtc/tree/v1.9.13#source-exec) source in go2rtc can be used for custom ffmpeg commands and other applications. An example is below:
 
 :::warning
 
@@ -244,16 +244,11 @@ The `exec:`, `echo:`, and `expr:` sources are disabled by default for security.
 
 :::
 
-:::warning
-
-The `exec:`, `echo:`, and `expr:` sources are disabled by default for security. You must set `GO2RTC_ALLOW_ARBITRARY_EXEC=true` to use them. See [Security: Restricted Stream Sources](#security-restricted-stream-sources) for more information.
-
-:::
-
-NOTE: The output will need to be passed with two curly braces `{{output}}`
+NOTE: RTSP output will need to be passed with two curly braces `{{output}}`, whereas pipe output must be passed without curly braces.
 
 ```yaml
 go2rtc:
   streams:
     stream1: exec:ffmpeg -hide_banner -re -stream_loop -1 -i /media/BigBuckBunny.mp4 -c copy -rtsp_transport tcp -f rtsp {{output}}
+    stream2: exec:rpicam-vid -t 0 --libav-format h264 -o -
 ```

docs/docs/frigate/hardware.md

@@ -223,10 +223,11 @@ Apple Silicon can not run within a container, so a ZMQ proxy is utilized to comm
 
 With the [ROCm](../configuration/object_detectors.md#amdrocm-gpu-detector) detector Frigate can take advantage of many discrete AMD GPUs.
 
-| Name      | YOLOv9 Inference Time       | YOLO-NAS Inference Time   |
-| --------- | --------------------------- | ------------------------- |
-| AMD 780M  | t-320: ~ 14 ms s-320: 20 ms | 320: ~ 25 ms 640: ~ 50 ms |
-| AMD 8700G |                             | 320: ~ 20 ms 640: ~ 40 ms |
+| Name           | YOLOv9 Inference Time       | YOLO-NAS Inference Time   | RF-DETR Inference Time |
+| -------------- | --------------------------- | ------------------------- | ---------------------- |
+| AMD 780M       | t-320: ~ 14 ms s-320: 20 ms | 320: ~ 25 ms 640: ~ 50 ms |                        |
+| AMD 8700G      |                             | 320: ~ 20 ms 640: ~ 40 ms |                        |
+| AMD 9060XT 16G | t-320: ~ 4 ms  s-320: 5 ms  | 320: ~ 6 ms               | Nano-320: ~ 90 ms      |
 
 ## Community Supported Detectors
 

docs/docs/frigate/installation.md

@@ -4,12 +4,15 @@ title: Installation
 ---
 
 import ShmCalculator from '@site/src/components/ShmCalculator'
+import DockerComposeGenerator from '@site/src/components/DockerComposeGenerator'
+import Tabs from '@theme/Tabs';
+import TabItem from '@theme/TabItem';
 
 Frigate is a Docker container that can be run on any Docker host including as a [Home Assistant App](https://www.home-assistant.io/apps/). Note that the Home Assistant App is **not** the same thing as the integration. The [integration](/integrations/home-assistant) is required to integrate Frigate into Home Assistant, whether you are running Frigate as a standalone Docker container or as a Home Assistant App.
 
 :::tip
 
-If you already have Frigate installed as a Home Assistant App, check out the [getting started guide](../guides/getting_started#configuring-frigate) to configure Frigate.
+If you already have Frigate installed as a Home Assistant App, check out the [getting started guide](../guides/getting_started.md#configuring-frigate) to configure Frigate.
 
 :::
 
@@ -286,7 +289,7 @@ The MemryX MX3 Accelerator is available in the M.2 2280 form factor (like an NVM
 
 #### Installation
 
-To get started with MX3 hardware setup for your system, refer to the [Hardware Setup Guide](https://developer.memryx.com/get_started/hardware_setup.html).
+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).
 
 Then follow these steps for installing the correct driver/runtime configuration:
 
@@ -295,6 +298,12 @@ Then follow these steps for installing the correct driver/runtime configuration:
 3. Run the script with `./user_installation.sh`
 4. **Restart your computer** to complete driver installation.
 
+:::warning
+
+For manual setup, use **MemryX SDK 2.1** only. Other SDK versions are not supported for this setup. See the [SDK 2.1 documentation](https://developer.memryx.com/2p1/index.html)
+
+:::
+
 #### Setup
 
 To set up Frigate, follow the default installation instructions, for example: `ghcr.io/blakeblackshear/frigate:stable`
@@ -468,6 +477,16 @@ Finally, configure [hardware object detection](/configuration/object_detectors#a
 
 Running through Docker with Docker Compose is the recommended install method.
 
+<Tabs>
+  <TabItem value="domestic" label="Docker Compose Generator" default>
+
+Generate a Frigate Docker Compose configuration based on your hardware and requirements.
+
+<DockerComposeGenerator/>
+
+
+  </TabItem>
+  <TabItem value="original" label="Example Docker Compose File">
 ```yaml
 services:
   frigate:
@@ -501,6 +520,10 @@ services:
     environment:
       FRIGATE_RTSP_PASSWORD: "password"
 ```
+  </TabItem>
+</Tabs>
+
+**Docker CLI**
 
 If you can't use Docker Compose, you can run the container with something similar to this:
 

docs/docs/guides/getting_started.md

@@ -192,7 +192,7 @@ cameras:
 
 ### Step 4: Configure detectors
 
-By default, Frigate will use a single CPU detector.
+By default, Frigate will use a single OpenVINO detector running on the CPU.
 
 In many cases, the integrated graphics on Intel CPUs provides sufficient performance for typical Frigate setups. If you have an Intel processor, you can follow the configuration below.
 

docs/docs/integrations/third_party_extensions.md

@@ -39,6 +39,10 @@ This is a fork (with fixed errors and new features) of [original Double Take](ht
 
 [Frigate telegram](https://github.com/OldTyT/frigate-telegram) makes it possible to send events from Frigate to Telegram. Events are sent as a message with a text description, video, and thumbnail.
 
+## [kiosk-monitor](https://github.com/extremeshok/kiosk-monitor)
+
+[kiosk-monitor](https://github.com/extremeshok/kiosk-monitor) is a Raspberry Pi watchdog that runs Chromium fullscreen on a Frigate dashboard (optionally with VLC on a second monitor for an RTSP camera stream), auto-restarts on frozen screens or unreachable URLs, and ships a Birdseye-aware Chromium helper that auto-sizes the grid to the display.
+
 ## [Periscope](https://github.com/maksz42/periscope)
 
 [Periscope](https://github.com/maksz42/periscope) is a lightweight Android app that turns old devices into live viewers for Frigate. It works on Android 2.2 and above, including Android TV. It supports authentication and HTTPS.

docs/docs/troubleshooting/faqs.md

@@ -111,26 +111,16 @@ TCP ensures that all data packets arrive in the correct order. This is crucial f
 
 You can still configure Frigate to use UDP by using ffmpeg input args or the preset `preset-rtsp-udp`. See the [ffmpeg presets](/configuration/ffmpeg_presets) documentation.
 
-### Frigate hangs on startup with a "probing detect stream" message in the logs
+### Frigate is slow to start up with a "probing detect stream" message in the logs
 
-On startup, Frigate probes each camera's detect stream with OpenCV to auto-detect its resolution. OpenCV's FFmpeg backend may attempt RTSP over UDP during this probe regardless of the `-rtsp_transport tcp` in your `input_args` or preset. For cameras that do not respond to UDP (common on some Reolink models and others behind firewalls that block UDP), the probe can hang indefinitely and block Frigate from finishing startup, or it can return zeroed-out dimensions that 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 System Metrics.
 
-There are two ways to avoid this:
+To skip the probe entirely and make startup instant, set `detect.width` and `detect.height` explicitly in your camera config:
 
-1. Set `detect.width` and `detect.height` explicitly in your camera config. When both are set, Frigate skips the auto-detect probe entirely:
-
-   ```yaml
-   cameras:
-     my_camera:
-       detect:
-         width: 1280
-         height: 720
-   ```
-
-2. Force OpenCV's FFmpeg backend to use TCP for RTSP by setting the environment variable on your Frigate container:
-
-   ```
-   OPENCV_FFMPEG_CAPTURE_OPTIONS=rtsp_transport;tcp
-   ```
-
-   This is a process-wide setting and applies to all cameras. If you have any cameras that require `preset-rtsp-udp`, use option 1 instead.
+```yaml
+cameras:
+  my_camera:
+    detect:
+      width: 1280
+      height: 720
+```

docs/package-lock.json

@@ -14,9 +14,11 @@
         "@docusaurus/theme-mermaid": "^3.7.0",
         "@inkeep/docusaurus": "^2.0.16",
         "@mdx-js/react": "^3.1.0",
+        "@types/js-yaml": "^4.0.9",
         "clsx": "^2.1.1",
         "docusaurus-plugin-openapi-docs": "^4.5.1",
         "docusaurus-theme-openapi-docs": "^4.5.1",
+        "js-yaml": "^4.1.1",
         "prism-react-renderer": "^2.4.1",
         "raw-loader": "^4.0.2",
         "react": "^18.3.1",
@@ -5747,6 +5749,11 @@
         "@types/istanbul-lib-report": "*"
       }
     },
+    "node_modules/@types/js-yaml": {
+      "version": "4.0.9",
+      "resolved": "https://mirrors.tencent.com/npm/@types/js-yaml/-/js-yaml-4.0.9.tgz",
+      "integrity": "sha512-k4MGaQl5TGo/iipqb2UDG2UwjXziSWkh0uysQelTlJpX1qGlpUZYm8PnO4DxG1qBomtJUdYJ6qR6xdIah10JLg=="
+    },
     "node_modules/@types/json-schema": {
       "version": "7.0.15",
       "resolved": "https://registry.npmjs.org/@types/json-schema/-/json-schema-7.0.15.tgz",
@@ -10897,9 +10904,9 @@
       "license": "MIT"
     },
     "node_modules/express/node_modules/path-to-regexp": {
-      "version": "0.1.12",
-      "resolved": "https://registry.npmjs.org/path-to-regexp/-/path-to-regexp-0.1.12.tgz",
-      "integrity": "sha512-RA1GjUVMnvYFxuqovrEqZoxxW5NUZqbwKtYz/Tt7nXerk0LbLblQmrsgdeOxV5SFHf0UDggjS/bSeOZwt1pmEQ==",
+      "version": "0.1.13",
+      "resolved": "https://registry.npmjs.org/path-to-regexp/-/path-to-regexp-0.1.13.tgz",
+      "integrity": "sha512-A/AGNMFN3c8bOlvV9RreMdrv7jsmF9XIfDeCd87+I8RNg6s78BhJxMu69NEMHBSJFxKidViTEdruRwEk/WIKqA==",
       "license": "MIT"
     },
     "node_modules/express/node_modules/range-parser": {
@@ -10964,9 +10971,9 @@
       "license": "MIT"
     },
     "node_modules/fast-uri": {
-      "version": "3.1.0",
-      "resolved": "https://registry.npmjs.org/fast-uri/-/fast-uri-3.1.0.tgz",
-      "integrity": "sha512-iPeeDKJSWf4IEOasVVrknXpaBV0IApz/gp7S2bb7Z4Lljbl2MGJRqInZiUrQwV16cpzw/D3S5j5Julj/gT52AA==",
+      "version": "3.1.2",
+      "resolved": "https://registry.npmjs.org/fast-uri/-/fast-uri-3.1.2.tgz",
+      "integrity": "sha512-rVjf7ArG3LTk+FS6Yw81V1DLuZl1bRbNrev6Tmd/9RaroeeRRJhAt7jg/6YFxbvAQXUCavSoZhPPj6oOx+5KjQ==",
       "funding": [
         {
           "type": "github",
@@ -12883,7 +12890,7 @@
     },
     "node_modules/js-yaml": {
       "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/js-yaml/-/js-yaml-4.1.1.tgz",
+      "resolved": "https://mirrors.tencent.com/npm/js-yaml/-/js-yaml-4.1.1.tgz",
       "integrity": "sha512-qQKT4zQxXl8lLwBtHMWwaTcGfFOZviOJet3Oy/xmGk2gZH677CJM9EvtfdSkgWcATZhj/55JZ0rmy3myCT5lsA==",
       "license": "MIT",
       "dependencies": {

docs/package.json

@@ -3,9 +3,10 @@
   "version": "0.0.0",
   "private": true,
   "scripts": {
+    "build:config": "node scripts/build-config.mjs",
     "docusaurus": "docusaurus",
-    "start": "npm run regen-docs && docusaurus start --host 0.0.0.0",
-    "build": "npm run regen-docs && docusaurus build",
+    "start": "npm run build:config && npm run regen-docs && docusaurus start --host 0.0.0.0",
+    "build": "npm run build:config && npm run regen-docs && docusaurus build",
     "swizzle": "docusaurus swizzle",
     "deploy": "docusaurus deploy",
     "clear": "docusaurus clear",
@@ -23,9 +24,11 @@
     "@docusaurus/theme-mermaid": "^3.7.0",
     "@inkeep/docusaurus": "^2.0.16",
     "@mdx-js/react": "^3.1.0",
+    "@types/js-yaml": "^4.0.9",
     "clsx": "^2.1.1",
     "docusaurus-plugin-openapi-docs": "^4.5.1",
     "docusaurus-theme-openapi-docs": "^4.5.1",
+    "js-yaml": "^4.1.1",
     "prism-react-renderer": "^2.4.1",
     "raw-loader": "^4.0.2",
     "react": "^18.3.1",

docs/scripts/build-config.mjs

@@ -0,0 +1,64 @@
+#!/usr/bin/env node
+
+/**
+ * Build script: reads config.yaml and generates TypeScript files
+ * for the Docker Compose Generator.
+ *
+ * Usage: node scripts/build-config.mjs
+ */
+
+import fs from "node:fs";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+import yaml from "js-yaml";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const CONFIG_DIR = path.resolve(__dirname, "../src/components/DockerComposeGenerator/config");
+const YAML_PATH = path.join(CONFIG_DIR, "config.yaml");
+
+// Read & parse YAML
+const raw = fs.readFileSync(YAML_PATH, "utf8");
+const config = yaml.load(raw);
+
+if (!config.devices || !config.hardware || !config.ports) {
+  console.error("config.yaml must contain 'devices', 'hardware', and 'ports' sections.");
+  process.exit(1);
+}
+
+/**
+ * Generate a .ts file from a section of the YAML config.
+ */
+function generateTsFile(sectionName, items, typeName, varName, mapVarName, yamlFilename) {
+  const jsonItems = JSON.stringify(items, null, 2);
+  // Indent JSON to fit inside the array literal
+  const indented = jsonItems
+    .split("\n")
+    .map((line, i) => (i === 0 ? line : "  " + line))
+    .join("\n");
+
+  const content = `/**
+ * AUTO-GENERATED FILE — do not edit directly.
+ * Source: ${yamlFilename}
+ * To update, edit the YAML file and run: npm run build:config
+ */
+
+import type { ${typeName} } from "./types";
+
+export const ${varName}: ${typeName}[] = ${indented};
+
+/** Lookup map for quick access by ID */
+export const ${mapVarName}: Map<string, ${typeName}> = new Map(${varName}.map((item) => [item.id, item]));
+`;
+
+  const outPath = path.join(CONFIG_DIR, `${sectionName}.ts`);
+  fs.writeFileSync(outPath, content, "utf8");
+  console.log(`  ✓ Generated ${sectionName}.ts (${items.length} items)`);
+}
+
+console.log("Building config from config.yaml...");
+
+generateTsFile("devices", config.devices, "DeviceConfig", "devices", "deviceMap", "config.yaml");
+generateTsFile("hardware", config.hardware, "HardwareOption", "hardwareOptions", "hardwareMap", "config.yaml");
+generateTsFile("ports", config.ports, "PortConfig", "ports", "portMap", "config.yaml");
+
+console.log("Done!");

docs/src/components/DockerComposeGenerator/DockerComposeGenerator.tsx

@@ -0,0 +1,108 @@
+import React from "react";
+import Admonition from "@theme/Admonition";
+import DeviceSelector from "./components/DeviceSelector";
+import HardwareOptions from "./components/HardwareOptions";
+import PortConfigSection from "./components/PortConfig";
+import StoragePaths from "./components/StoragePaths";
+import NvidiaGpuConfig from "./components/NvidiaGpuConfig";
+import OtherOptions from "./components/OtherOptions";
+import GeneratedOutput from "./components/GeneratedOutput";
+import { useConfigGenerator } from "./hooks/useConfigGenerator";
+import styles from "./styles.module.css";
+
+/**
+ * Simple markdown-link-to-React renderer for help text.
+ * Only supports [text](url) syntax — no nested brackets.
+ */
+function renderHelpText(text: string): React.ReactNode {
+  const parts = text.split(/(\[[^\]]+\]\([^)]+\))/g);
+  return parts.map((part, i) => {
+    const match = part.match(/^\[([^\]]+)\]\(([^)]+)\)$/);
+    if (match) {
+      return (
+        <a key={i} href={match[2]}>
+          {match[1]}
+        </a>
+      );
+    }
+    return <React.Fragment key={i}>{part}</React.Fragment>;
+  });
+}
+
+export default function DockerComposeGenerator() {
+  const {
+    deviceId, device, hardwareEnabled,
+    portEnabled,
+    nvidiaGpuCount, nvidiaGpuDeviceId,
+    configPath, mediaPath, rtspPassword, timezone, shmSize,
+    shmSizeError, gpuDeviceIdError, configPathError, mediaPathError,
+    hasAnyHardware, generatedYaml,
+    selectDevice, toggleHardware, togglePort,
+    handleShmSizeChange, handleConfigPathChange, handleMediaPathChange,
+    handleNvidiaGpuCountChange, handleNvidiaGpuDeviceIdChange,
+    setRtspPassword, setTimezone, isHardwareDisabled,
+  } = useConfigGenerator();
+
+  return (
+    <div className={styles.generator}>
+      <div className={styles.card}>
+        <DeviceSelector selectedId={deviceId} onSelect={selectDevice} />
+
+        {device.helpText && (
+          <Admonition type={device.helpType || "info"}>
+            {renderHelpText(device.helpText)}
+          </Admonition>
+        )}
+
+        {device.needsNvidiaConfig && (
+          <NvidiaGpuConfig
+            gpuCount={nvidiaGpuCount}
+            gpuDeviceId={nvidiaGpuDeviceId}
+            gpuDeviceIdError={gpuDeviceIdError}
+            onGpuCountChange={handleNvidiaGpuCountChange}
+            onGpuDeviceIdChange={handleNvidiaGpuDeviceIdChange}
+          />
+        )}
+
+        <HardwareOptions
+          deviceId={deviceId}
+          hardwareEnabled={hardwareEnabled}
+          onToggle={toggleHardware}
+          isDisabled={isHardwareDisabled}
+        />
+
+        <StoragePaths
+          configPath={configPath}
+          mediaPath={mediaPath}
+          configPathError={configPathError}
+          mediaPathError={mediaPathError}
+          onConfigPathChange={handleConfigPathChange}
+          onMediaPathChange={handleMediaPathChange}
+        />
+
+        <PortConfigSection
+          portEnabled={portEnabled}
+          onTogglePort={togglePort}
+        />
+
+        <OtherOptions
+          rtspPassword={rtspPassword}
+          timezone={timezone}
+          shmSize={shmSize}
+          shmSizeError={shmSizeError}
+          onRtspPasswordChange={setRtspPassword}
+          onTimezoneChange={setTimezone}
+          onShmSizeChange={handleShmSizeChange}
+        />
+
+        <GeneratedOutput
+          yaml={generatedYaml}
+          configPath={configPath}
+          mediaPath={mediaPath}
+          hasAnyHardware={hasAnyHardware}
+          deviceId={deviceId}
+        />
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/DeviceSelector.tsx

@@ -0,0 +1,147 @@
+import React from "react";
+import { useColorMode } from "@docusaurus/theme-common";
+import { devices } from "../config";
+import type { DeviceConfig } from "../config";
+import styles from "../styles.module.css";
+
+interface Props {
+  selectedId: string;
+  onSelect: (id: string) => void;
+}
+
+/**
+ * Determine the icon type from the icon string:
+ * - Starts with "<svg" → inline SVG
+ * - Starts with "/" or "http" → image URL/path
+ * - Otherwise → emoji text
+ */
+function getIconType(icon: string): "svg" | "image" | "emoji" {
+  const trimmed = icon.trim();
+  if (trimmed.startsWith("<svg")) return "svg";
+  if (trimmed.startsWith("/") || trimmed.startsWith("http://") || trimmed.startsWith("https://")) return "image";
+  return "emoji";
+}
+
+/**
+ * Check if the style object contains background-* properties,
+ * indicating the image should be rendered as a CSS background-image
+ * rather than an <img> tag.
+ */
+function hasBackgroundProps(style: React.CSSProperties | undefined): boolean {
+  if (!style) return false;
+  return Object.keys(style).some((key) => {
+    const k = key.toLowerCase().replace(/-/g, "");
+    return k === "backgroundsize" || k === "backgroundposition" || k === "backgroundrepeat" || k === "backgroundimage";
+  });
+}
+
+/**
+ * Convert a style object to CSS custom properties (e.g. { width: "24px" } → { "--svg-width": "24px" })
+ * so they can be consumed by CSS rules targeting child elements like <svg>.
+ */
+function toCssVars(style: React.CSSProperties | undefined, prefix: string): React.CSSProperties {
+  if (!style) return {};
+  const vars: Record<string, string> = {};
+  for (const [key, value] of Object.entries(style)) {
+    const cssKey = key.replace(/([A-Z])/g, "-$1").toLowerCase();
+    vars[`--${prefix}-${cssKey}`] = value;
+  }
+  return vars as React.CSSProperties;
+}
+
+function DeviceIcon({ device }: { device: DeviceConfig }) {
+  const { isDarkTheme } = useColorMode();
+  const iconStr = isDarkTheme && device.iconDark ? device.iconDark : device.icon;
+  const iconStyle = (isDarkTheme && device.iconDarkStyle
+    ? device.iconDarkStyle
+    : device.iconStyle) as React.CSSProperties | undefined;
+  const svgStyle = (isDarkTheme && device.svgDarkStyle
+    ? device.svgDarkStyle
+    : device.svgStyle) as React.CSSProperties | undefined;
+
+  const iconType = getIconType(iconStr);
+
+  if (iconType === "svg") {
+    return (
+      <div
+        className={styles.deviceIconSvg}
+        style={{ ...iconStyle, ...toCssVars(svgStyle, "svg") }}
+        dangerouslySetInnerHTML={{ __html: iconStr }}
+      />
+    );
+  }
+
+  if (iconType === "image") {
+    // When iconStyle contains background-* properties, render as background-image
+    // on the container div instead of an <img> tag, enabling background-size/position control.
+    if (hasBackgroundProps(iconStyle)) {
+      return (
+        <div
+          className={styles.deviceIconImage}
+          style={{
+            backgroundImage: `url(${iconStr})`,
+            backgroundRepeat: "no-repeat",
+            backgroundPosition: "center",
+            backgroundSize: "contain",
+            ...iconStyle,
+          }}
+        />
+      );
+    }
+    return (
+      <div className={styles.deviceIconImage}>
+        <img src={iconStr} alt={device.name} style={iconStyle} />
+      </div>
+    );
+  }
+
+  return (
+    <div className={styles.deviceIcon} style={iconStyle}>
+      {iconStr}
+    </div>
+  );
+}
+
+function DeviceCard({
+  device,
+  active,
+  onClick,
+}: {
+  device: DeviceConfig;
+  active: boolean;
+  onClick: () => void;
+}) {
+  return (
+    <div
+      className={`${styles.deviceCard} ${active ? styles.deviceCardActive : ""}`}
+      onClick={onClick}
+      role="button"
+      tabIndex={0}
+      onKeyDown={(e) => {
+        if (e.key === "Enter" || e.key === " ") onClick();
+      }}
+    >
+      <DeviceIcon device={device} />
+      <div className={styles.deviceName}>{device.name}</div>
+      <div className={styles.deviceDesc}>{device.description}</div>
+    </div>
+  );
+}
+
+export default function DeviceSelector({ selectedId, onSelect }: Props) {
+  return (
+    <div className={styles.formSection}>
+      <h4>Device Type</h4>
+      <div className={styles.deviceGrid}>
+        {devices.map((d) => (
+          <DeviceCard
+            key={d.id}
+            device={d}
+            active={selectedId === d.id}
+            onClick={() => onSelect(d.id)}
+          />
+        ))}
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/GeneratedOutput.tsx

@@ -0,0 +1,60 @@
+import React, { useState, useCallback } from "react";
+import CodeBlock from "@theme/CodeBlock";
+import Admonition from "@theme/Admonition";
+import styles from "../styles.module.css";
+
+interface Props {
+  yaml: string;
+  configPath: string;
+  mediaPath: string;
+  hasAnyHardware: boolean;
+  deviceId: string;
+}
+
+export default function GeneratedOutput({
+  yaml,
+  configPath,
+  mediaPath,
+  hasAnyHardware,
+  deviceId,
+}: Props) {
+  const [copied, setCopied] = useState(false);
+
+  const handleCopy = useCallback(() => {
+    navigator.clipboard.writeText(yaml).then(() => {
+      setCopied(true);
+      setTimeout(() => setCopied(false), 2000);
+    });
+  }, [yaml]);
+
+  return (
+    <div className={styles.resultSection}>
+      <div className={styles.resultHeader}>
+        <h4>Generated Configuration</h4>
+        <button className="button button--primary button--sm" onClick={handleCopy}>
+          {copied ? "Copied!" : "Copy"}
+        </button>
+      </div>
+
+      {!configPath && (
+        <Admonition type="tip">
+          <p>You haven&apos;t specified a config file directory. You may want to modify the default path.</p>
+        </Admonition>
+      )}
+      {!mediaPath && (
+        <Admonition type="tip">
+          <p>You haven&apos;t specified a recording storage directory. You may want to modify the default path.</p>
+        </Admonition>
+      )}
+      {deviceId === "stable" && !hasAnyHardware && (
+        <Admonition type="warning">
+          <p>You haven&apos;t selected any hardware acceleration. Please check if you have supported hardware available.</p>
+        </Admonition>
+      )}
+
+      <CodeBlock language="yaml" title="docker-compose.yml">
+        {yaml}
+      </CodeBlock>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/HardwareOptions.tsx

@@ -0,0 +1,62 @@
+import React from "react";
+import { hardwareOptions } from "../config";
+import type { HardwareOption } from "../config";
+import styles from "../styles.module.css";
+
+interface Props {
+  deviceId: string;
+  hardwareEnabled: Record<string, boolean>;
+  onToggle: (hwId: string) => void;
+  isDisabled: (hwId: string) => boolean;
+}
+
+function renderDescription(text: string): React.ReactNode {
+  const parts = text.split(/(\[[^\]]+\]\([^)]+\))/g);
+  return parts.map((part, i) => {
+    const match = part.match(/^\[([^\]]+)\]\(([^)]+)\)$/);
+    if (match) {
+      return <a key={i} href={match[2]}>{match[1]}</a>;
+    }
+    return <React.Fragment key={i}>{part}</React.Fragment>;
+  });
+}
+
+function HardwareCheckbox({
+  hw, disabled, checked, onToggle,
+}: {
+  hw: HardwareOption; disabled: boolean; checked: boolean; onToggle: () => void;
+}) {
+  return (
+    <div className={styles.hardwareItem}>
+      <label className={`${styles.checkboxLabel} ${disabled ? styles.checkboxDisabled : ""}`}>
+        <input type="checkbox" checked={checked} onChange={onToggle} disabled={disabled} />
+        <span>{hw.label}</span>
+      </label>
+      {checked && hw.description && (
+        <div className={styles.hardwareDescription}>{renderDescription(hw.description)}</div>
+      )}
+    </div>
+  );
+}
+
+export default function HardwareOptions({ deviceId, hardwareEnabled, onToggle, isDisabled }: Props) {
+  return (
+    <div className={styles.formSection}>
+      <h4>Generic Hardware Devices</h4>
+      {deviceId !== "stable" && (
+        <p className={styles.helpText}>
+          Some options have been auto-configured based on your device type.
+        </p>
+      )}
+      <div className={styles.checkboxGrid}>
+        {hardwareOptions.map((hw) => {
+          const disabled = isDisabled(hw.id);
+          const checked = disabled ? false : !!hardwareEnabled[hw.id];
+          return (
+            <HardwareCheckbox key={hw.id} hw={hw} disabled={disabled} checked={checked} onToggle={() => onToggle(hw.id)} />
+          );
+        })}
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/NvidiaGpuConfig.tsx

@@ -0,0 +1,64 @@
+import React from "react";
+import styles from "../styles.module.css";
+
+interface Props {
+  gpuCount: string;
+  gpuDeviceId: string;
+  gpuDeviceIdError: boolean;
+  onGpuCountChange: (value: string) => void;
+  onGpuDeviceIdChange: (value: string) => void;
+}
+
+export default function NvidiaGpuConfig({
+  gpuCount,
+  gpuDeviceId,
+  gpuDeviceIdError,
+  onGpuCountChange,
+  onGpuDeviceIdChange,
+}: Props) {
+  const showDeviceId = gpuCount !== "";
+
+  return (
+    <div className={styles.nvidiaConfig}>
+      <div className={styles.formGroup}>
+        <label htmlFor="dcg-gpu-count" className={styles.label}>
+          GPU count:
+        </label>
+        <input
+          id="dcg-gpu-count"
+          type="text"
+          inputMode="numeric"
+          pattern="[0-9]*"
+          className={styles.input}
+          value={gpuCount}
+          placeholder="all"
+          onChange={(e) => onGpuCountChange(e.target.value.replace(/\D/g, ""))}
+        />
+      </div>
+      {showDeviceId && (
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-gpu-device-id" className={styles.label}>
+            GPU device IDs (required, comma-separated):
+          </label>
+          <input
+            id="dcg-gpu-device-id"
+            type="text"
+            className={`${styles.input} ${gpuDeviceIdError ? styles.inputError : ""}`}
+            value={gpuDeviceId}
+            placeholder="0"
+            onChange={(e) => onGpuDeviceIdChange(e.target.value)}
+          />
+          {gpuDeviceIdError ? (
+            <p className={styles.helpText}>
+              ⚠️ GPU device IDs are required when GPU count is a number
+            </p>
+          ) : (
+            <p className={styles.helpText}>
+              Single GPU: 0 &nbsp;|&nbsp; Multiple GPUs: 0,1,2
+            </p>
+          )}
+        </div>
+      )}
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/OtherOptions.tsx

@@ -0,0 +1,122 @@
+import React, { useMemo } from "react";
+import CodeInline from "@theme/CodeInline";
+import styles from "../styles.module.css";
+
+const AUTO_TIMEZONE_VALUE = "__auto__";
+
+function getTimezoneList(): string[] {
+  if (typeof Intl !== "undefined") {
+    const intl = Intl as typeof Intl & {
+      supportedValuesOf?: (key: string) => string[];
+    };
+    const supported = intl.supportedValuesOf?.("timeZone");
+    if (supported && supported.length > 0) {
+      return [...supported].sort();
+    }
+  }
+
+  const fallback = Intl.DateTimeFormat().resolvedOptions().timeZone;
+  return fallback ? [fallback] : ["UTC"];
+}
+
+interface Props {
+  rtspPassword: string;
+  timezone: string;
+  shmSize: string;
+  shmSizeError: boolean;
+  onRtspPasswordChange: (value: string) => void;
+  onTimezoneChange: (value: string) => void;
+  onShmSizeChange: (value: string) => void;
+}
+
+export default function OtherOptions({
+  rtspPassword,
+  timezone,
+  shmSize,
+  shmSizeError,
+  onRtspPasswordChange,
+  onTimezoneChange,
+  onShmSizeChange,
+}: Props) {
+  const timezones = useMemo(() => getTimezoneList(), []);
+  const systemTimezone =
+    Intl.DateTimeFormat().resolvedOptions().timeZone || "Etc/UTC";
+  const selectedValue = timezone || AUTO_TIMEZONE_VALUE;
+
+  return (
+    <div className={styles.formSection}>
+      <h4>Other Options</h4>
+      <div className={styles.formGrid}>
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-timezone" className={styles.label}>
+            Timezone:
+          </label>
+          <select
+            id="dcg-timezone"
+            className={`${styles.input} ${styles.select}`}
+            value={selectedValue}
+            onChange={(e) =>
+              onTimezoneChange(
+                e.target.value === AUTO_TIMEZONE_VALUE ? "" : e.target.value
+              )
+            }
+          >
+            <option value={AUTO_TIMEZONE_VALUE}>
+              Use browser timezone ({systemTimezone})
+            </option>
+            {timezones.map((tz) => (
+              <option key={tz} value={tz}>
+                {tz}
+              </option>
+            ))}
+          </select>
+        </div>
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-shm-size" className={styles.label}>
+            Shared memory (SHM):
+          </label>
+          <input
+            id="dcg-shm-size"
+            type="text"
+            className={`${styles.input} ${shmSizeError ? styles.inputError : ""}`}
+            value={shmSize}
+            placeholder="512mb"
+            onChange={(e) => onShmSizeChange(e.target.value)}
+          />
+          {shmSizeError ? (
+            <p className={styles.helpText}>
+              ⚠️ Invalid format. Use a number followed by a unit (e.g. 512mb, 1gb)
+            </p>
+          ) : (
+            <p className={styles.helpText}>
+              See{" "}
+              <a href="/frigate/installation#calculating-required-shm-size">
+                calculating required SHM size
+              </a>{" "}
+              for the correct value.
+            </p>
+          )}
+        </div>
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-rtsp-password" className={styles.label}>
+            RTSP password:
+          </label>
+          <input
+            id="dcg-rtsp-password"
+            type="text"
+            className={styles.input}
+            value={rtspPassword}
+            placeholder="password"
+            onChange={(e) => onRtspPasswordChange(e.target.value)}
+          />
+          <p className={styles.helpText}>
+            Optional. You can specify{" "}
+            <CodeInline>{"{FRIGATE_RTSP_PASSWORD}"}</CodeInline>{" "}
+            in the config file to reference camera stream passwords. This is NOT
+            the Frigate login password.
+          </p>
+        </div>
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/PortConfig.tsx

@@ -0,0 +1,71 @@
+import React from "react";
+import Admonition from "@theme/Admonition";
+import { ports } from "../config";
+import styles from "../styles.module.css";
+
+interface Props {
+  portEnabled: Record<string, boolean>;
+  onTogglePort: (portId: string) => void;
+}
+
+function PortItem({
+  port,
+  enabled,
+  onToggle,
+}: {
+  port: typeof ports[number];
+  enabled: boolean;
+  onToggle: () => void;
+}) {
+  const showWarning = port.warningContent && (
+    port.warningWhen === "checked" ? enabled :
+    port.warningWhen === "unchecked" ? !enabled : enabled
+  );
+
+  return (
+    <div className={styles.hardwareItem}>
+      <label className={`${styles.checkboxLabel} ${port.locked ? styles.checkboxDisabled : ""}`}>
+        <input
+          type="checkbox"
+          checked={enabled}
+          onChange={onToggle}
+          disabled={port.locked}
+        />
+        <span>
+          {port.locked && "🔒 "}
+          Port {port.host}
+          {port.protocol !== "tcp" && `/${port.protocol}`}
+        </span>
+      </label>
+      {port.description && (
+        <div className={styles.hardwareDescription}>{port.description}</div>
+      )}
+      {showWarning && (
+        <Admonition type={port.warningType || "warning"}>
+          {port.warningContent}
+        </Admonition>
+      )}
+    </div>
+  );
+}
+
+export default function PortConfigSection({
+  portEnabled,
+  onTogglePort,
+}: Props) {
+  return (
+    <div className={styles.formSection}>
+      <h4>Port Configuration</h4>
+      <div className={styles.checkboxGrid}>
+        {ports.map((port) => (
+          <PortItem
+            key={port.id}
+            port={port}
+            enabled={!!portEnabled[port.id]}
+            onToggle={() => onTogglePort(port.id)}
+          />
+        ))}
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/components/StoragePaths.tsx

@@ -0,0 +1,66 @@
+import React from "react";
+import styles from "../styles.module.css";
+
+interface Props {
+  configPath: string;
+  mediaPath: string;
+  configPathError: boolean;
+  mediaPathError: boolean;
+  onConfigPathChange: (value: string) => void;
+  onMediaPathChange: (value: string) => void;
+}
+
+export default function StoragePaths({
+  configPath,
+  mediaPath,
+  configPathError,
+  mediaPathError,
+  onConfigPathChange,
+  onMediaPathChange,
+}: Props) {
+  return (
+    <div className={styles.formSection}>
+      <h4>Storage Paths</h4>
+      <div className={styles.formGrid}>
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-config-path" className={styles.label}>
+            Config / DB / model cache directory (on your host):
+          </label>
+          <input
+            id="dcg-config-path"
+            type="text"
+            className={`${styles.input} ${configPathError ? styles.inputError : ""}`}
+            value={configPath}
+            placeholder="/path/to/your/config"
+            onChange={(e) => onConfigPathChange(e.target.value)}
+          />
+          {configPathError && (
+            <p className={styles.helpText}>
+              ⚠️ Path contains invalid characters. Only letters, numbers,
+              underscores, hyphens, slashes, and dots are allowed.
+            </p>
+          )}
+        </div>
+        <div className={styles.formGroup}>
+          <label htmlFor="dcg-media-path" className={styles.label}>
+            Recording storage directory (on your host):
+          </label>
+          <input
+            id="dcg-media-path"
+            type="text"
+            className={`${styles.input} ${mediaPathError ? styles.inputError : ""}`}
+            value={mediaPath}
+            placeholder="/path/to/your/storage"
+            onChange={(e) => onMediaPathChange(e.target.value)}
+          />
+          {mediaPathError && (
+            <p className={styles.helpText}>
+              ⚠️ Path contains invalid characters. Only letters, numbers,
+              underscores, hyphens, slashes, and dots are allowed.
+            </p>
+          )}
+        </div>
+      </div>
+    </div>
+  );
+}

docs/src/components/DockerComposeGenerator/config/index.ts

@@ -0,0 +1,12 @@
+export { devices, deviceMap } from "./devices";
+export { hardwareOptions, hardwareMap } from "./hardware";
+export { ports, portMap } from "./ports";
+
+export type {
+  DeviceConfig,
+  DeviceMapping,
+  VolumeMapping,
+  HardwareOption,
+  PortConfig,
+  NvidiaDeployConfig,
+} from "./types";

docs/src/components/DockerComposeGenerator/config/types.ts

@@ -0,0 +1,154 @@
+/**
+ * Type definitions for the Docker Compose Generator configuration.
+ * All device, hardware, and port options are declaratively defined
+ * so that adding a new device only requires editing config files.
+ */
+
+/** A single device mapping entry (e.g. /dev/dri:/dev/dri) */
+export interface DeviceMapping {
+  /** Host device path */
+  host: string;
+  /** Container device path (defaults to host if omitted) */
+  container?: string;
+  /** Inline comment for this device line */
+  comment?: string;
+}
+
+/** A single volume mapping entry */
+export interface VolumeMapping {
+  /** Host path */
+  host: string;
+  /** Container path */
+  container: string;
+  /** Whether the mount is read-only */
+  readOnly?: boolean;
+  /** Inline comment */
+  comment?: string;
+}
+
+/** NVIDIA deploy configuration for docker-compose */
+export interface NvidiaDeployConfig {
+  /** "all" or a specific number */
+  count: string;
+  /** Specific GPU device IDs (when count is a number) */
+  deviceIds?: string[];
+}
+
+/** Full device type definition */
+export interface DeviceConfig {
+  /** Unique identifier, e.g. "intel" */
+  id: string;
+  /** Display name, e.g. "Intel GPU" */
+  name: string;
+  /** Short description */
+  description: string;
+  /**
+   * Icon for the device card. Supports:
+   * - Emoji string (e.g. "🖥️")
+   * - Image URL or static path (e.g. "/img/intel.svg", "https://example.com/icon.png")
+   * - Inline SVG markup (e.g. "<svg>...</svg>")
+   */
+  icon: string;
+  /**
+   * Additional CSS properties applied to the icon element.
+   * - For image-type icons: if any `background-*` property (e.g. `background-size`,
+   *   `background-position`) is present, the image is rendered as a CSS `background-image`
+   *   on the container div, enabling full background positioning control.
+   *   Otherwise the image is rendered as an `<img>` tag and styles apply to it.
+   * - For emoji/SVG icons: styles apply to the container div.
+   */
+  iconStyle?: Record<string, string>;
+  /**
+   * Additional CSS properties applied directly to the inner `<svg>` element
+   * when the icon is an inline SVG. Use this to override the default
+   * `width: 100%; height: 100%` or set `fill`, `transform`, etc.
+   * Ignored for emoji and image-type icons.
+   */
+  svgStyle?: Record<string, string>;
+  /**
+   * Icon for dark mode. Same format as `icon`. When provided, this icon
+   * replaces `icon` when the user is in dark mode.
+   */
+  iconDark?: string;
+  /** Additional CSS properties for the dark mode icon container */
+  iconDarkStyle?: Record<string, string>;
+  /**
+   * SVG-specific styles for dark mode. Same as `svgStyle` but applied
+   * when dark mode is active. Merged over `svgStyle` in dark mode.
+   */
+  svgDarkStyle?: Record<string, string>;
+  /** Docker image tag, e.g. "stable" */
+  imageTag: string;
+  /**
+   * Image tag suffix appended to the base tag.
+   * e.g. "-standard-arm64" produces "stable-standard-arm64"
+   */
+  imageTagSuffix?: string;
+  /** Hardware option IDs to auto-enable when this device is selected */
+  autoHardware: string[];
+  /** Help text shown as an admonition when this device is selected */
+  helpText?: string;
+  /** Admonition type for help text */
+  helpType?: "info" | "warning" | "danger";
+  /** Device mappings always added for this device type */
+  devices?: DeviceMapping[];
+  /** Volume mappings always added for this device type */
+  volumes?: VolumeMapping[];
+  /** Extra environment variables for this device type */
+  env?: Record<string, string>;
+  /** NVIDIA deploy config (only for tensorrt) */
+  nvidiaDeploy?: NvidiaDeployConfig;
+  /** Runtime setting, e.g. "nvidia" for Jetson */
+  runtime?: string;
+  /** Extra hosts entries, e.g. "host.docker.internal:host-gateway" */
+  extraHosts?: string[];
+  /** Security options, e.g. ["apparmor=unconfined"] */
+  securityOpt?: string[];
+  /** Whether this device type needs the NVIDIA GPU config UI */
+  needsNvidiaConfig?: boolean;
+}
+
+/** Generic hardware acceleration option definition */
+export interface HardwareOption {
+  /** Unique identifier, e.g. "usbCoral" */
+  id: string;
+  /** Display label */
+  label: string;
+  /**
+   * Description shown below the checkbox when this option is enabled.
+   * Supports markdown link syntax: [text](url)
+   */
+  description?: string;
+  /** Device IDs that disable this option */
+  disabledWhen?: string[];
+  /** Device mappings added when this option is enabled */
+  devices?: DeviceMapping[];
+  /** Volume mappings added when this option is enabled */
+  volumes?: VolumeMapping[];
+  /** Extra environment variables */
+  env?: Record<string, string>;
+}
+
+/** Port definition */
+export interface PortConfig {
+  /** Unique identifier (also the default host port as string) */
+  id: string;
+  /** Host port number */
+  host: number;
+  /** Container port number */
+  container: number;
+  /** Protocol */
+  protocol?: "tcp" | "udp";
+  /** Description of the port's purpose */
+  description: string;
+  /** Whether enabled by default */
+  defaultEnabled: boolean;
+  /** Whether this port is locked (always enabled, cannot be toggled off) */
+  locked?: boolean;
+  /** Admonition type for the warning */
+  warningType?: "warning" | "danger";
+  /** Warning content (markdown) */
+  warningContent?: string;
+  /** When to show the warning: when the port is checked or unchecked */
+  warningWhen?: "checked" | "unchecked";
+}

docs/src/components/DockerComposeGenerator/generator/index.ts

@@ -0,0 +1,250 @@
+import type {
+  DeviceConfig,
+  DeviceMapping,
+  VolumeMapping,
+} from "../config/types";
+import { hardwareMap } from "../config";
+
+// ---------------------------------------------------------------------------
+// Input type
+// ---------------------------------------------------------------------------
+
+export interface GeneratorInput {
+  device: DeviceConfig;
+  selectedHardware: string[];
+  enabledPorts: string[];
+  configPath: string;
+  mediaPath: string;
+  rtspPassword?: string;
+  timezone: string;
+  shmSize: string;
+  nvidiaGpuCount?: string;
+  nvidiaGpuDeviceId?: string;
+}
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function deviceLine(dm: DeviceMapping): string {
+  const host = dm.host;
+  const container = dm.container ?? dm.host;
+  const mapping = host === container ? host : `${host}:${container}`;
+  const comment = dm.comment ? ` # ${dm.comment}` : "";
+  return `      - ${mapping}${comment}`;
+}
+
+function volumeLine(vm: VolumeMapping): string {
+  const ro = vm.readOnly ? ":ro" : "";
+  const comment = vm.comment ? ` # ${vm.comment}` : "";
+  return `      - ${vm.host}:${vm.container}${ro}${comment}`;
+}
+
+// ---------------------------------------------------------------------------
+// YAML builder — each section returns an array of lines
+// ---------------------------------------------------------------------------
+
+function buildImage(device: DeviceConfig): string[] {
+  const tag = device.imageTagSuffix
+    ? `${device.imageTag}${device.imageTagSuffix}`
+    : device.imageTag;
+  return [`    image: ghcr.io/blakeblackshear/frigate:${tag}`];
+}
+
+function buildDevices(
+  device: DeviceConfig,
+  hwDevices: DeviceMapping[]
+): string[] {
+  const all: DeviceMapping[] = [
+    ...(device.devices ?? []),
+    ...hwDevices,
+  ];
+  if (all.length === 0) return [];
+  return [
+    "    devices:",
+    ...all.map(deviceLine),
+  ];
+}
+
+function buildVolumes(
+  device: DeviceConfig,
+  hwVolumes: VolumeMapping[],
+  configPath: string,
+  mediaPath: string
+): string[] {
+  const all: VolumeMapping[] = [
+    ...(device.volumes ?? []),
+    ...hwVolumes,
+  ];
+  return [
+    "    volumes:",
+    "      - /etc/localtime:/etc/localtime:ro # Sync host time",
+    `      - ${configPath}:/config # Config file directory`,
+    `      - ${mediaPath}:/media/frigate # Recording storage directory`,
+    "      - type: tmpfs # 1GB in-memory filesystem for recording segment storage",
+    "        target: /tmp/cache",
+    "        tmpfs:",
+    "          size: 1000000000",
+    ...all.map(volumeLine),
+  ];
+}
+
+function buildPorts(enabledPorts: string[]): string[] {
+  return [
+    "    ports:",
+    ...enabledPorts,
+  ];
+}
+
+function buildEnvironment(
+  device: DeviceConfig,
+  hwEnv: Record<string, string>,
+  rtspPassword: string | undefined,
+  timezone: string
+): string[] {
+  const allEnv: Record<string, string> = {
+    ...hwEnv,
+    ...(device.env ?? {}),
+  };
+
+  const lines: string[] = ["    environment:"];
+
+  if (rtspPassword) {
+    lines.push(
+      `      FRIGATE_RTSP_PASSWORD: "${rtspPassword}" # RTSP password — change to your own`
+    );
+  }
+
+  lines.push(`      TZ: "${timezone}" # Timezone`);
+
+  for (const [key, value] of Object.entries(allEnv)) {
+    lines.push(`      ${key}: "${value}"`);
+  }
+
+  return lines;
+}
+
+function buildDeploy(device: DeviceConfig, input: GeneratorInput): string[] {
+  if (device.id === "stable-tensorrt") {
+    const count = input.nvidiaGpuCount || "all";
+    const isAll = count === "all";
+    const deviceId = input.nvidiaGpuDeviceId?.trim();
+
+    if (isAll) {
+      return [
+        "    deploy:",
+        "      resources:",
+        "        reservations:",
+        "          devices:",
+        "            - driver: nvidia",
+        "              count: all # Use all GPUs",
+        "              capabilities: [gpu]",
+      ];
+    }
+
+    if (deviceId) {
+      const ids = deviceId
+        .split(",")
+        .map((s) => s.trim())
+        .filter(Boolean)
+        .map((s) => `'${s}'`)
+        .join(", ");
+      return [
+        "    deploy:",
+        "      resources:",
+        "        reservations:",
+        "          devices:",
+        "            - driver: nvidia",
+        `              device_ids: [${ids}] # GPU device IDs`,
+        `              count: ${count} # GPU count`,
+        "              capabilities: [gpu]",
+      ];
+    }
+
+    return [
+      "    deploy:",
+      "      resources:",
+      "        reservations:",
+      "          devices:",
+      "            - driver: nvidia",
+      `              count: ${count} # GPU count`,
+      "              capabilities: [gpu]",
+    ];
+  }
+
+  return [];
+}
+
+function buildRuntime(device: DeviceConfig): string[] {
+  if (device.runtime) {
+    return [`    runtime: ${device.runtime}`];
+  }
+  return [];
+}
+
+function buildExtraHosts(device: DeviceConfig): string[] {
+  if (!device.extraHosts?.length) return [];
+  return [
+    "    extra_hosts:",
+    ...device.extraHosts.map(
+      (h, i) =>
+        `      - "${h}"${i === 0 ? " # Required to talk to the NPU detector" : ""}`
+    ),
+  ];
+}
+
+function buildSecurityOpt(device: DeviceConfig): string[] {
+  if (!device.securityOpt?.length) return [];
+  return [
+    "    security_opt:",
+    ...device.securityOpt.map((s) => `      - ${s}`),
+  ];
+}
+
+// ---------------------------------------------------------------------------
+// Public API
+// ---------------------------------------------------------------------------
+
+/**
+ * Generate a docker-compose YAML string from the given input.
+ * The output is pure YAML with inline comments (no Shiki annotations).
+ */
+export function generateDockerCompose(input: GeneratorInput): string {
+  const { device } = input;
+
+  // Collect hardware-level devices, volumes, and env
+  const hwDevices: DeviceMapping[] = [];
+  const hwVolumes: VolumeMapping[] = [];
+  const hwEnv: Record<string, string> = {};
+
+  for (const hwId of input.selectedHardware) {
+    const hw = hardwareMap.get(hwId);
+    if (!hw) continue;
+    // Skip GPU device mapping for tensorrt images (it uses deploy instead)
+    if (hw.id === "gpu" && device.imageTag === "stable-tensorrt") continue;
+    hwDevices.push(...(hw.devices ?? []));
+    hwVolumes.push(...(hw.volumes ?? []));
+    Object.assign(hwEnv, hw.env ?? {});
+  }
+
+  const lines: string[] = [
+    "services:",
+    "  frigate:",
+    "    container_name: frigate",
+    "    privileged: true # This may not be necessary for all setups",
+    "    restart: unless-stopped",
+    "    stop_grace_period: 30s # Allow enough time to shut down the various services",
+    ...buildImage(device),
+    `    shm_size: "${input.shmSize || "512mb"}" # Update for your cameras based on SHM calculation`,
+    ...buildRuntime(device),
+    ...buildDeploy(device, input),
+    ...buildExtraHosts(device),
+    ...buildSecurityOpt(device),
+    ...buildDevices(device, hwDevices),
+    ...buildVolumes(device, hwVolumes, input.configPath, input.mediaPath),
+    ...buildPorts(input.enabledPorts),
+    ...buildEnvironment(device, hwEnv, input.rtspPassword, input.timezone),
+  ];
+
+  return lines.join("\n");
+}

docs/src/components/DockerComposeGenerator/hooks/useConfigGenerator.ts

@@ -0,0 +1,195 @@
+import { useState, useCallback, useMemo } from "react";
+import { deviceMap, hardwareMap, portMap } from "../config";
+import { generateDockerCompose } from "../generator";
+import type { GeneratorInput } from "../generator";
+
+/**
+ * Main hook that holds all form state and generates the Docker Compose output.
+ * Configuration is loaded synchronously from build-time generated .ts files.
+ */
+export function useConfigGenerator() {
+  const [deviceId, setDeviceId] = useState("stable");
+
+  const [hardwareEnabled, setHardwareEnabled] = useState<Record<string, boolean>>(() => {
+    const defaultDevice = deviceMap.get("stable");
+    const initial: Record<string, boolean> = {};
+    if (defaultDevice) {
+      for (const hwId of defaultDevice.autoHardware) {
+        initial[hwId] = true;
+      }
+    }
+    return initial;
+  });
+
+  const [portEnabled, setPortEnabled] = useState<Record<string, boolean>>(() => {
+    const initial: Record<string, boolean> = {};
+    for (const p of portMap.values()) {
+      initial[p.id] = p.defaultEnabled;
+    }
+    return initial;
+  });
+
+  const [nvidiaGpuCount, setNvidiaGpuCount] = useState("");
+  const [nvidiaGpuDeviceId, setNvidiaGpuDeviceId] = useState("");
+  const [configPath, setConfigPath] = useState("");
+  const [mediaPath, setMediaPath] = useState("");
+  const [rtspPassword, setRtspPassword] = useState("");
+  const [timezone, setTimezone] = useState("");
+  const [shmSize, setShmSize] = useState("512mb");
+  const [shmSizeError, setShmSizeError] = useState(false);
+  const [gpuDeviceIdError, setGpuDeviceIdError] = useState(false);
+  const [configPathError, setConfigPathError] = useState(false);
+  const [mediaPathError, setMediaPathError] = useState(false);
+
+  const device = useMemo(() => deviceMap.get(deviceId)!, [deviceId]);
+
+  const selectDevice = useCallback((id: string) => {
+    const newDevice = deviceMap.get(id);
+    if (!newDevice) return;
+    setDeviceId(id);
+    setHardwareEnabled(() => {
+      const next: Record<string, boolean> = {};
+      for (const hwId of newDevice.autoHardware) {
+        next[hwId] = true;
+      }
+      return next;
+    });
+    setNvidiaGpuCount("");
+    setNvidiaGpuDeviceId("");
+    setGpuDeviceIdError(false);
+  }, []);
+
+  const toggleHardware = useCallback((hwId: string) => {
+    setHardwareEnabled((prev) => ({ ...prev, [hwId]: !prev[hwId] }));
+  }, []);
+
+  const togglePort = useCallback((portId: string) => {
+    const port = portMap.get(portId);
+    if (port?.locked) return;
+    setPortEnabled((prev) => ({ ...prev, [portId]: !prev[portId] }));
+  }, []);
+
+  const isHardwareDisabled = useCallback(
+    (hwId: string): boolean => {
+      const hw = hardwareMap.get(hwId);
+      if (!hw) return false;
+      return hw.disabledWhen?.includes(deviceId) ?? false;
+    },
+    [deviceId]
+  );
+
+  const validateShmSize = useCallback((value: string): boolean => {
+    if (!value) return true;
+    return /^\d+(\.\d+)?[bkmgBKMG]{1,2}$/.test(value);
+  }, []);
+
+  const validatePath = useCallback((value: string): boolean => {
+    if (!value) return true;
+    return /^[a-zA-Z0-9_\-/./]+$/.test(value);
+  }, []);
+
+  const handleShmSizeChange = useCallback(
+    (value: string) => {
+      const filtered = value.replace(/[^0-9.bkmgBKMG]/g, "");
+      const valid = validateShmSize(filtered);
+      setShmSize(filtered);
+      setShmSizeError(!valid && filtered !== "");
+    },
+    [validateShmSize]
+  );
+
+  const handleConfigPathChange = useCallback(
+    (value: string) => {
+      const filtered = value.replace(/[^a-zA-Z0-9_\-/./]/g, "");
+      const valid = validatePath(filtered);
+      setConfigPath(filtered);
+      setConfigPathError(!valid && filtered !== "");
+    },
+    [validatePath]
+  );
+
+  const handleMediaPathChange = useCallback(
+    (value: string) => {
+      const filtered = value.replace(/[^a-zA-Z0-9_\-/./]/g, "");
+      const valid = validatePath(filtered);
+      setMediaPath(filtered);
+      setMediaPathError(!valid && filtered !== "");
+    },
+    [validatePath]
+  );
+
+  const handleNvidiaGpuCountChange = useCallback((value: string) => {
+    // Only allow digits
+    setNvidiaGpuCount(value);
+    if (value === "") {
+      setNvidiaGpuDeviceId("");
+      setGpuDeviceIdError(false);
+    } else {
+      setGpuDeviceIdError(false);
+    }
+  }, []);
+
+  const handleNvidiaGpuDeviceIdChange = useCallback((value: string) => {
+    setNvidiaGpuDeviceId(value.trim());
+    setGpuDeviceIdError(false);
+  }, []);
+
+  const enabledPortLines = useMemo(() => {
+    const lines: string[] = [];
+    for (const [id, enabled] of Object.entries(portEnabled)) {
+      if (!enabled) continue;
+      const p = portMap.get(id);
+      if (!p) continue;
+      const proto = p.protocol && p.protocol !== "tcp" ? `/${p.protocol}` : "";
+      const comment = p.description ? ` # ${p.description}` : "";
+      lines.push(`      - "${p.host}:${p.container}${proto}"${comment}`);
+    }
+    return lines;
+  }, [portEnabled]);
+
+  const selectedHardwareIds = useMemo(() => {
+    return Object.entries(hardwareEnabled)
+      .filter(([id, enabled]) => {
+        if (!enabled) return false;
+        const hw = hardwareMap.get(id);
+        if (!hw) return false;
+        if (hw.disabledWhen?.includes(deviceId)) return false;
+        return true;
+      })
+      .map(([id]) => id);
+  }, [hardwareEnabled, deviceId]);
+
+  const generatedYaml = useMemo(() => {
+    const input: GeneratorInput = {
+      device,
+      selectedHardware: selectedHardwareIds,
+      enabledPorts: enabledPortLines,
+      configPath: configPath || "/path/to/your/config",
+      mediaPath: mediaPath || "/path/to/your/storage",
+      rtspPassword,
+      timezone: timezone || Intl.DateTimeFormat().resolvedOptions().timeZone || "Etc/UTC",
+      shmSize: shmSize || "512mb",
+      nvidiaGpuCount,
+      nvidiaGpuDeviceId,
+    };
+    return generateDockerCompose(input);
+  }, [
+    device, selectedHardwareIds, enabledPortLines,
+    configPath, mediaPath, rtspPassword, timezone, shmSize,
+    nvidiaGpuCount, nvidiaGpuDeviceId,
+  ]);
+
+  const hasAnyHardware = selectedHardwareIds.length > 0 || !!device?.devices?.length;
+
+  return {
+    deviceId, device, hardwareEnabled, portEnabled,
+    nvidiaGpuCount, nvidiaGpuDeviceId,
+    configPath, mediaPath, rtspPassword, timezone, shmSize,
+    shmSizeError, gpuDeviceIdError, configPathError, mediaPathError,
+    hasAnyHardware, generatedYaml,
+    selectDevice, toggleHardware, togglePort,
+    handleShmSizeChange, handleConfigPathChange, handleMediaPathChange,
+    handleNvidiaGpuCountChange, handleNvidiaGpuDeviceIdChange,
+    setRtspPassword, setTimezone, isHardwareDisabled,
+  };
+}

docs/src/components/DockerComposeGenerator/index.ts

@@ -0,0 +1 @@
+export { default } from "./DockerComposeGenerator";

docs/src/components/DockerComposeGenerator/styles.module.css

@@ -0,0 +1,381 @@
+/* ===================================================================
+   Docker Compose Generator — styles
+   Uses Docusaurus / Infima CSS variables for theme compatibility.
+   =================================================================== */
+
+.generator {
+  margin: 2rem 0;
+}
+
+.card {
+  background: var(--ifm-background-surface-color);
+  border: 1px solid var(--ifm-color-emphasis-400);
+  border-radius: 12px;
+  padding: 2rem;
+  box-shadow: var(--ifm-global-shadow-lw);
+}
+
+[data-theme="light"] .card {
+  background: var(--ifm-color-emphasis-100);
+  border: 1px solid var(--ifm-color-emphasis-300);
+}
+
+/* --- Form sections --- */
+
+.formSection {
+  margin-bottom: 1.5rem;
+  padding-bottom: 1.5rem;
+  border-bottom: 1px solid var(--ifm-color-emphasis-400);
+}
+
+.formSection:last-child {
+  border-bottom: none;
+  margin-bottom: 0;
+  padding-bottom: 0;
+}
+
+.formSection h4 {
+  margin: 0 0 1rem 0;
+  color: var(--ifm-font-color-base);
+  font-size: 1.1rem;
+  font-weight: var(--ifm-font-weight-semibold);
+}
+
+/* --- Form controls --- */
+
+.formGroup {
+  margin-bottom: 1rem;
+}
+
+.formGroup:last-child {
+  margin-bottom: 0;
+}
+
+.label {
+  display: block;
+  margin-bottom: 0.25rem;
+  color: var(--ifm-font-color-base);
+  font-weight: var(--ifm-font-weight-semibold);
+  font-size: 0.9rem;
+}
+
+.input {
+  width: 100%;
+  padding: 0.5rem 0.75rem;
+  border: 1px solid var(--ifm-color-emphasis-400);
+  border-radius: 6px;
+  background: var(--ifm-background-color);
+  color: var(--ifm-font-color-base);
+  font-size: 0.95rem;
+  transition: border-color 0.2s, box-shadow 0.2s;
+}
+
+[data-theme="light"] .input {
+  background: #fff;
+  border: 1px solid #d0d7de;
+}
+
+.input:focus {
+  outline: none;
+  border-color: var(--ifm-color-primary);
+  box-shadow: 0 0 0 3px var(--ifm-color-primary-lightest);
+}
+
+[data-theme="dark"] .input {
+  border-color: var(--ifm-color-emphasis-300);
+}
+
+.inputError {
+  border-color: #e74c3c;
+  animation: shake 0.3s ease-in-out;
+}
+
+@keyframes shake {
+  0%,
+  100% {
+    transform: translateX(0);
+  }
+  25% {
+    transform: translateX(-5px);
+  }
+  75% {
+    transform: translateX(5px);
+  }
+}
+
+/* --- Select dropdown --- */
+
+.select {
+  cursor: pointer;
+  appearance: none;
+  -moz-appearance: none;
+  -webkit-appearance: none;
+  background: var(--ifm-background-color)
+    url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='12' viewBox='0 0 12 12'%3E%3Cpath fill='%23666' d='M6 8L1 3h10z'/%3E%3C/svg%3E")
+    no-repeat right 0.75rem center / 12px 12px;
+  padding-right: 2rem;
+}
+
+[data-theme="light"] .select {
+  background: #fff
+    url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='12' viewBox='0 0 12 12'%3E%3Cpath fill='%23555' d='M6 8L1 3h10z'/%3E%3C/svg%3E")
+    no-repeat right 0.75rem center / 12px 12px;
+}
+
+.helpText {
+  margin: 0.5rem 0 0 0;
+  font-size: 0.85rem;
+  color: var(--ifm-font-color-secondary);
+  line-height: 1.5;
+}
+
+.helpText a {
+  color: var(--ifm-color-primary);
+}
+
+/* --- Device grid --- */
+
+.deviceGrid {
+  display: grid;
+  grid-template-columns: repeat(auto-fill, minmax(130px, 1fr));
+  gap: 0.75rem;
+  margin-top: 0.5rem;
+}
+
+.deviceCard {
+  padding: 0.75rem;
+  border: 2px solid var(--ifm-color-emphasis-400);
+  border-radius: 12px;
+  cursor: pointer;
+  transition: all 0.2s;
+  text-align: center;
+  background: var(--ifm-background-color);
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+}
+
+[data-theme="light"] .deviceCard {
+  border: 2px solid #d0d7de;
+  background: #fff;
+}
+
+.deviceCard:hover {
+  border-color: var(--ifm-color-primary);
+  background: var(--ifm-color-emphasis-100);
+  transform: translateY(-2px);
+}
+
+.deviceCardActive {
+  border-color: var(--ifm-color-primary);
+  background: var(--ifm-color-primary-lightest);
+  box-shadow: 0 0 0 1px var(--ifm-color-primary);
+}
+
+[data-theme="light"] .deviceCardActive {
+  background: color-mix(in srgb, var(--ifm-color-primary) 12%, #fff);
+}
+
+[data-theme="dark"] .deviceCardActive {
+  background: color-mix(in srgb, var(--ifm-color-primary) 25%, #1b1b1b);
+}
+
+[data-theme="dark"] .deviceCardActive .deviceName {
+  color: var(--ifm-color-primary-light);
+}
+
+[data-theme="dark"] .deviceCardActive .deviceDesc {
+  color: var(--ifm-color-primary-light);
+  opacity: 0.85;
+}
+
+.deviceIcon {
+  font-size: 2rem;
+  margin-bottom: 0.25rem;
+  height: 40px;
+  width: 50px;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+}
+
+.deviceIconSvg {
+  margin-bottom: 0.25rem;
+  height: 40px;
+  width: 50px;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  overflow: visible;
+  /* Allow iconStyle width/height to override */
+  flex-shrink: 0;
+}
+
+.deviceIconSvg svg {
+  width: var(--svg-width, 100%);
+  height: var(--svg-height, 100%);
+  fill: var(--svg-fill, currentColor);
+  transform: var(--svg-transform, none);
+}
+
+.deviceIconImage {
+  margin-bottom: 0.25rem;
+  height: 40px;
+  width: 50px;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+}
+
+.deviceIconImage img {
+  max-width: 100%;
+  max-height: 100%;
+  object-fit: contain;
+}
+
+.deviceName {
+  font-weight: var(--ifm-font-weight-semibold);
+  color: var(--ifm-font-color-base);
+  margin-bottom: 0.15rem;
+  font-size: 0.9rem;
+}
+
+.deviceDesc {
+  font-size: 0.75rem;
+  color: var(--ifm-font-color-secondary);
+  line-height: 1.3;
+}
+
+/* --- Checkbox grid --- */
+
+.checkboxGrid {
+  display: grid;
+  grid-template-columns: repeat(2, 1fr);
+  gap: 0.5rem;
+}
+
+@media (max-width: 576px) {
+  .checkboxGrid {
+    grid-template-columns: 1fr;
+  }
+}
+
+.hardwareItem {
+  margin-bottom: 0;
+}
+
+.hardwareDescription {
+  margin: 0.15rem 0 0.4rem 1.6rem;
+  font-size: 0.8rem;
+  color: var(--ifm-font-color-secondary);
+  line-height: 1.5;
+}
+
+.hardwareDescription a {
+  color: var(--ifm-color-primary);
+  text-decoration: underline;
+  text-underline-offset: 2px;
+}
+
+.checkboxLabel {
+  display: flex;
+  align-items: center;
+  gap: 0.5rem;
+  cursor: pointer;
+  padding: 0.4rem 0.5rem;
+  border-radius: 6px;
+  transition: background-color 0.2s;
+  font-size: 0.9rem;
+}
+
+.checkboxLabel:hover {
+  background: var(--ifm-color-emphasis-100);
+}
+
+.checkboxLabel input[type="checkbox"] {
+  width: 1.1rem;
+  height: 1.1rem;
+  cursor: pointer;
+  flex-shrink: 0;
+}
+
+.checkboxLabel span {
+  color: var(--ifm-font-color-base);
+}
+
+.checkboxDisabled {
+  cursor: not-allowed;
+}
+
+.checkboxDisabled:hover {
+  background: transparent;
+}
+
+.checkboxDisabled input[type="checkbox"] {
+  cursor: not-allowed;
+  opacity: 0.5;
+}
+
+/* --- Form grid (side-by-side) --- */
+
+.formGrid {
+  display: grid;
+  grid-template-columns: repeat(2, 1fr);
+  gap: 1rem;
+}
+
+@media (max-width: 576px) {
+  .formGrid {
+    grid-template-columns: 1fr;
+  }
+}
+
+.formGrid .formGroup {
+  margin-bottom: 0;
+}
+
+/* --- Port section --- */
+
+.portSection {
+  margin-bottom: 0.75rem;
+}
+
+.warningBadge {
+  margin-left: auto;
+  color: #e67e22;
+  font-size: 0.85rem;
+}
+
+/* --- NVIDIA config --- */
+
+.nvidiaConfig {
+  margin-top: 1rem;
+  margin-bottom: 1.5rem;
+  padding: 1rem;
+  background: var(--ifm-background-color);
+  border-radius: 8px;
+  border-left: 3px solid var(--ifm-color-primary);
+}
+
+[data-theme="light"] .nvidiaConfig {
+  background: #f6f8fa;
+  border-left: 3px solid var(--ifm-color-primary);
+}
+
+/* --- Result section --- */
+
+.resultSection {
+  margin-top: 2rem;
+}
+
+.resultHeader {
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  margin-bottom: 1rem;
+}
+
+.resultHeader h4 {
+  margin: 0;
+  color: var(--ifm-font-color-base);
+}

docs/static/frigate-api.yaml

@@ -5997,7 +5997,10 @@ paths:
       tags:
         - App
       summary: Start debug replay
-      description: Start a debug replay session from camera recordings.
+      description:
+        Start a debug replay session from camera recordings. Returns
+        immediately while clip generation runs as a background job; subscribe
+        to the 'debug_replay' job_state WS topic to track progress.
       operationId: start_debug_replay_debug_replay_start_post
       requestBody:
         required: true
@@ -6006,12 +6009,16 @@ paths:
             schema:
               $ref: "#/components/schemas/DebugReplayStartBody"
       responses:
-        "200":
+        "202":
           description: Successful Response
           content:
             application/json:
               schema:
                 $ref: "#/components/schemas/DebugReplayStartResponse"
+        "400":
+          description: Invalid camera, time range, or no recordings
+        "409":
+          description: A replay session is already active
         "422":
           description: Validation Error
           content:
@@ -6272,10 +6279,14 @@ components:
         replay_camera:
           type: string
           title: Replay Camera
+        job_id:
+          type: string
+          title: Job Id
       type: object
       required:
         - success
         - replay_camera
+        - job_id
       title: DebugReplayStartResponse
       description: Response for starting a debug replay session.
     DebugReplayStatusResponse:

frigate/api/app.py

@@ -96,11 +96,46 @@ def version():
 
 
 @router.get("/stats", dependencies=[Depends(allow_any_authenticated())])
-def stats(request: Request):
-    return JSONResponse(content=request.app.stats_emitter.get_latest_stats())
+def stats(
+    request: Request,
+    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+):
+    stats_data = request.app.stats_emitter.get_latest_stats()
+
+    # Admins see the full snapshot
+    if request.headers.get("remote-role") == "admin":
+        return JSONResponse(content=stats_data)
+
+    allowed_set = set(allowed_cameras)
+
+    # Shallow-copy so we don't mutate the cached stats history entry.
+    filtered = {**stats_data}
+
+    cameras = stats_data.get("cameras")
+    if cameras is not None:
+        filtered["cameras"] = {
+            name: data for name, data in cameras.items() if name in allowed_set
+        }
+
+    bandwidth = stats_data.get("bandwidth_usages")
+    if bandwidth is not None:
+        filtered["bandwidth_usages"] = {
+            name: data for name, data in bandwidth.items() if name in allowed_set
+        }
+
+    # cmdline can leak camera URLs/paths; strip but keep cpu/mem so
+    # client-side problem heuristics still work.
+    cpu_usages = stats_data.get("cpu_usages")
+    if cpu_usages is not None:
+        filtered["cpu_usages"] = {
+            pid: {k: v for k, v in usage.items() if k != "cmdline"}
+            for pid, usage in cpu_usages.items()
+        }
+
+    return JSONResponse(content=filtered)
 
 
-@router.get("/stats/history", dependencies=[Depends(allow_any_authenticated())])
+@router.get("/stats/history", dependencies=[Depends(require_role(["admin"]))])
 def stats_history(request: Request, keys: str = None):
     if keys:
         keys = keys.split(",")
@@ -146,8 +181,13 @@ def config(request: Request):
         for name, detector in config_obj.detectors.items()
     }
 
-    # remove the mqtt password
+    # remove environment_vars for non-admin users
+    if request.headers.get("remote-role") != "admin":
+        config.pop("environment_vars", None)
+
+    # remove mqtt credentials
     config["mqtt"].pop("password", None)
+    config["mqtt"].pop("user", None)
 
     # remove the proxy secret
     config["proxy"].pop("auth_secret", None)
@@ -494,6 +534,40 @@ def config_save(save_option: str, body: Any = Body(media_type="text/plain")):
         )
 
 
+def _restore_masked_camera_paths(config_data: dict, config: FrigateConfig) -> None:
+    """Substitute incoming `*:*` masked credentials with the in-memory ones.
+
+    The /config response masks ffmpeg input credentials, so the settings UI
+    sends the masked path back when sibling fields (e.g. hwaccel_args) are
+    edited.  Without this we'd write `rtsp://*:*@host` into YAML and lose
+    the real credentials.  Mutates `config_data` in place.
+    """
+    cameras = config_data.get("cameras")
+    if not isinstance(cameras, dict):
+        return
+
+    for camera_name, camera_data in cameras.items():
+        if not isinstance(camera_data, dict):
+            continue
+        inputs = camera_data.get("ffmpeg", {}).get("inputs")
+        if not isinstance(inputs, list):
+            continue
+        existing = config.cameras.get(camera_name)
+        if existing is None:
+            continue
+        existing_paths = [inp.path for inp in existing.ffmpeg.inputs]
+        for index, input_obj in enumerate(inputs):
+            if not isinstance(input_obj, dict):
+                continue
+            path = input_obj.get("path")
+            if not isinstance(path, str):
+                continue
+            if ("://*:*@" in path or "user=*&password=*" in path) and index < len(
+                existing_paths
+            ):
+                input_obj["path"] = existing_paths[index]
+
+
 def _config_set_in_memory(request: Request, body: AppConfigSetBody) -> JSONResponse:
     """Apply config changes in-memory only, without writing to YAML.
 
@@ -504,6 +578,7 @@ def _config_set_in_memory(request: Request, body: AppConfigSetBody) -> JSONRespo
     try:
         updates = {}
         if body.config_data:
+            _restore_masked_camera_paths(body.config_data, request.app.frigate_config)
             updates = flatten_config_data(body.config_data)
             updates = {k: ("" if v is None else v) for k, v in updates.items()}
 
@@ -610,6 +685,9 @@ def config_set(request: Request, body: AppConfigSetBody):
                 if query_string:
                     updates = process_config_query_string(query_string)
                 elif body.config_data:
+                    _restore_masked_camera_paths(
+                        body.config_data, request.app.frigate_config
+                    )
                     updates = flatten_config_data(body.config_data)
                     # Convert None values to empty strings for deletion (e.g., when deleting masks)
                     updates = {k: ("" if v is None else v) for k, v in updates.items()}
@@ -792,7 +870,7 @@ def nvinfo():
 @router.get(
     "/logs/{service}",
     tags=[Tags.logs],
-    dependencies=[Depends(allow_any_authenticated())],
+    dependencies=[Depends(require_role(["admin"]))],
 )
 async def logs(
     service: str = Path(enum=["frigate", "nginx", "go2rtc"]),
@@ -997,12 +1075,27 @@ def get_media_sync_status(job_id: str):
 
 
 @router.get("/labels", dependencies=[Depends(allow_any_authenticated())])
-def get_labels(camera: str = ""):
+def get_labels(
+    camera: str = "",
+    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+):
     try:
         if camera:
+            if camera not in allowed_cameras:
+                return JSONResponse(
+                    content={
+                        "success": False,
+                        "message": f"Access denied to camera '{camera}'",
+                    },
+                    status_code=403,
+                )
             events = Event.select(Event.label).where(Event.camera == camera).distinct()
         else:
-            events = Event.select(Event.label).distinct()
+            events = (
+                Event.select(Event.label)
+                .where(Event.camera << allowed_cameras)
+                .distinct()
+            )
     except Exception as e:
         logger.error(e)
         return JSONResponse(
@@ -1015,9 +1108,16 @@ def get_labels(camera: str = ""):
 
 
 @router.get("/sub_labels", dependencies=[Depends(allow_any_authenticated())])
-def get_sub_labels(split_joined: Optional[int] = None):
+def get_sub_labels(
+    split_joined: Optional[int] = None,
+    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+):
     try:
-        events = Event.select(Event.sub_label).distinct()
+        events = (
+            Event.select(Event.sub_label)
+            .where(Event.camera << allowed_cameras)
+            .distinct()
+        )
     except Exception:
         return JSONResponse(
             content=({"success": False, "message": "Failed to get sub_labels"}),

frigate/api/auth.py

@@ -26,6 +26,7 @@ from frigate.api.defs.request.app_body import (
     AppPutRoleBody,
 )
 from frigate.api.defs.tags import Tags
+from frigate.api.media_auth import check_camera_access, deny_response_for_media_uri
 from frigate.config import AuthConfig, NetworkingConfig, ProxyConfig
 from frigate.const import CONFIG_DIR, JWT_SECRET_ENV_VAR, PASSWORD_HASH_ALGORITHM
 from frigate.models import User
@@ -633,6 +634,9 @@ def auth(request: Request):
         logger.debug("X-Proxy-Secret header does not match configured secret value")
         return fail_response
 
+    original_url = request.headers.get("x-original-url")
+    frigate_config = request.app.frigate_config
+
     # if auth is disabled, just apply the proxy header map and return success
     if not auth_config.enabled:
         # pass the user header value from the upstream proxy if a mapping is specified
@@ -649,6 +653,11 @@ def auth(request: Request):
         role = resolve_role(request.headers, proxy_config, config_roles_set)
 
         success_response.headers["remote-role"] = role
+
+        deny_status = deny_response_for_media_uri(original_url, role, frigate_config)
+        if deny_status is not None:
+            return Response("", status_code=deny_status)
+
         return success_response
 
     # now apply authentication
@@ -743,6 +752,11 @@ def auth(request: Request):
 
         success_response.headers["remote-user"] = user
         success_response.headers["remote-role"] = role
+
+        deny_status = deny_response_for_media_uri(original_url, role, frigate_config)
+        if deny_status is not None:
+            return Response("", status_code=deny_status)
+
         return success_response
     except Exception as e:
         logger.error(f"Error parsing jwt: {e}")
@@ -812,6 +826,11 @@ limiter = Limiter(key_func=get_remote_addr)
 )
 @limiter.limit(limit_value=rateLimiter.get_limit)
 def login(request: Request, body: AppPostLoginBody):
+    if not request.app.frigate_config.auth.enabled:
+        return JSONResponse(
+            content={"message": "Authentication is disabled"}, status_code=404
+        )
+
     JWT_COOKIE_NAME = request.app.frigate_config.auth.cookie_name
     JWT_COOKIE_SECURE = request.app.frigate_config.auth.cookie_secure
     JWT_SESSION_LENGTH = request.app.frigate_config.auth.session_length
@@ -1064,19 +1083,19 @@ async def require_camera_access(
         raise HTTPException(status_code=current_user.status_code, detail=detail)
 
     role = current_user["role"]
-    all_camera_names = set(request.app.frigate_config.cameras.keys())
-    roles_dict = request.app.frigate_config.auth.roles
-    allowed_cameras = User.get_allowed_cameras(role, roles_dict, all_camera_names)
+    frigate_config = request.app.frigate_config
 
-    # Admin or full access bypasses
-    if role == "admin" or not roles_dict.get(role):
+    if check_camera_access(role, camera_name, frigate_config):
         return
 
-    if camera_name not in allowed_cameras:
-        raise HTTPException(
-            status_code=403,
-            detail=f"Access denied to camera '{camera_name}'. Allowed: {allowed_cameras}",
-        )
+    all_camera_names = set(frigate_config.cameras.keys())
+    allowed_cameras = User.get_allowed_cameras(
+        role, frigate_config.auth.roles, all_camera_names
+    )
+    raise HTTPException(
+        status_code=403,
+        detail=f"Access denied to camera '{camera_name}'. Allowed: {allowed_cameras}",
+    )
 
 
 def _get_stream_owner_cameras(request: Request, stream_name: str) -> set[str]:

frigate/api/camera.py

@@ -19,7 +19,9 @@ from zeep.exceptions import Fault, TransportError
 from zeep.transports import AsyncTransport
 
 from frigate.api.auth import (
+    _get_stream_owner_cameras,
     allow_any_authenticated,
+    get_current_user,
     require_go2rtc_stream_access,
     require_role,
 )
@@ -31,11 +33,12 @@ from frigate.config.camera.updater import (
     CameraConfigUpdateTopic,
 )
 from frigate.config.env import substitute_frigate_vars
+from frigate.models import User
 from frigate.util.builtin import clean_camera_user_pass
 from frigate.util.camera_cleanup import cleanup_camera_db, cleanup_camera_files
 from frigate.util.config import find_config_file
 from frigate.util.image import run_ffmpeg_snapshot
-from frigate.util.services import ffprobe_stream
+from frigate.util.services import ffprobe_stream, is_restricted_go2rtc_source
 
 logger = logging.getLogger(__name__)
 
@@ -66,7 +69,7 @@ def _is_valid_host(host: str) -> bool:
 
 
 @router.get("/go2rtc/streams", dependencies=[Depends(allow_any_authenticated())])
-def go2rtc_streams():
+async def go2rtc_streams(request: Request):
     r = requests.get("http://127.0.0.1:1984/api/streams")
     if not r.ok:
         logger.error("Failed to fetch streams from go2rtc")
@@ -75,6 +78,24 @@ def go2rtc_streams():
             status_code=500,
         )
     stream_data = r.json()
+
+    # Roles with an explicit camera list see only streams owned by an allowed
+    # camera. Admin and full-access roles (no list / empty list) see all streams.
+    current_user = await get_current_user(request)
+    if not isinstance(current_user, JSONResponse):
+        role = current_user["role"]
+        roles_dict = request.app.frigate_config.auth.roles
+        if role != "admin" and roles_dict.get(role):
+            all_camera_names = set(request.app.frigate_config.cameras.keys())
+            allowed_cameras = set(
+                User.get_allowed_cameras(role, roles_dict, all_camera_names)
+            )
+            stream_data = {
+                name: data
+                for name, data in stream_data.items()
+                if _get_stream_owner_cameras(request, name) & allowed_cameras
+            }
+
     for data in stream_data.values():
         for producer in data.get("producers") or []:
             producer["url"] = clean_camera_user_pass(producer.get("url", ""))
@@ -126,9 +147,24 @@ def go2rtc_add_stream(request: Request, stream_name: str, src: str = ""):
         params = {"name": stream_name}
         if src:
             try:
-                params["src"] = substitute_frigate_vars(src)
+                resolved_src = substitute_frigate_vars(src)
             except KeyError:
-                params["src"] = src
+                resolved_src = src
+
+            if is_restricted_go2rtc_source(resolved_src):
+                logger.warning(
+                    "Rejected go2rtc stream '%s' with restricted source type (echo/expr/exec)",
+                    stream_name,
+                )
+                return JSONResponse(
+                    content={
+                        "success": False,
+                        "message": "Restricted stream source type",
+                    },
+                    status_code=400,
+                )
+
+            params["src"] = resolved_src
 
         r = requests.put(
             "http://127.0.0.1:1984/api/streams",
@@ -966,7 +1002,6 @@ async def onvif_probe(
                         probe = ffprobe_stream(
                             request.app.frigate_config.ffmpeg, test_uri, detailed=False
                         )
-                        print(probe)
                         ok = probe is not None and getattr(probe, "returncode", 1) == 0
                         tested_candidates.append(
                             {

frigate/api/chat.py

@@ -10,7 +10,7 @@ from functools import reduce
 from typing import Any, Dict, List, Optional
 
 import cv2
-from fastapi import APIRouter, Body, Depends, Request
+from fastapi import APIRouter, Body, Depends, HTTPException, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel
 
@@ -36,6 +36,8 @@ from frigate.api.defs.response.chat_response import (
 )
 from frigate.api.defs.tags import Tags
 from frigate.api.event import events
+from frigate.config import FrigateConfig
+from frigate.config.ui import UnitSystemEnum
 from frigate.genai.utils import build_assistant_message_for_conversation
 from frigate.jobs.vlm_watch import (
     get_vlm_watch_job,
@@ -66,62 +68,123 @@ class VLMMonitorRequest(BaseModel):
     zones: List[str] = []
 
 
-def get_tool_definitions() -> List[Dict[str, Any]]:
+def get_tool_definitions(
+    semantic_search_enabled: bool = False,
+) -> List[Dict[str, Any]]:
     """
     Get OpenAI-compatible tool definitions for Frigate.
 
     Returns a list of tool definitions that can be used with OpenAI-compatible
-    function calling APIs.
+    function calling APIs. When semantic search is enabled, the search_objects
+    tool exposes an additional `semantic_query` parameter for descriptive
+    queries (e.g. "person riding a lawn mower") and find_similar_objects is
+    included.
     """
+    search_objects_properties: Dict[str, Any] = {
+        "camera": {
+            "type": "string",
+            "description": "Camera name to filter by (optional).",
+        },
+        "label": {
+            "type": "string",
+            "description": (
+                "Generic object class to filter by — one of the tracked detector "
+                "labels such as 'person', 'package', 'car', 'dog', 'bird'. Use "
+                "this for broad queries like 'show me all cars today'. Combine "
+                "with semantic_query when the user also describes appearance or "
+                "behavior (e.g. label='person', semantic_query='riding a lawn "
+                "mower')."
+            ),
+        },
+        "sub_label": {
+            "type": "string",
+            "description": (
+                "Filter by a DISCRETE NAMED entity recognized in the detection. "
+                "Use this for: a known person's name ('John'), a delivery "
+                "company ('Amazon', 'UPS'), a recognized animal species or "
+                "breed ('blue jay', 'cardinal', 'golden retriever'), or a "
+                "license plate string. When filtering by a specific name, set "
+                "only sub_label and leave label unset. Do NOT use sub_label "
+                "for descriptions of appearance, clothing, or actions — those "
+                "belong in semantic_query."
+            ),
+        },
+        "after": {
+            "type": "string",
+            "description": "Start time in ISO 8601 format (e.g., '2024-01-01T00:00:00Z').",
+        },
+        "before": {
+            "type": "string",
+            "description": "End time in ISO 8601 format (e.g., '2024-01-01T23:59:59Z').",
+        },
+        "zones": {
+            "type": "array",
+            "items": {"type": "string"},
+            "description": "List of zone names to filter by.",
+        },
+        "limit": {
+            "type": "integer",
+            "description": "Maximum number of objects to return (default: 25).",
+            "default": 25,
+        },
+    }
+
+    if semantic_search_enabled:
+        search_objects_properties["semantic_query"] = {
+            "type": "string",
+            "description": (
+                "Optional natural-language description of a PHYSICAL "
+                "CHARACTERISTIC, APPEARANCE, or ACTIVITY the user mentioned, "
+                "used to semantically narrow results. Only set this when the "
+                "user describes something beyond what label and sub_label can "
+                "express on their own.\n"
+                "USE for descriptive phrases like: 'riding a lawn mower', "
+                "'wearing a red jacket', 'carrying a package', 'walking a "
+                "dog', 'on a bicycle', 'holding an umbrella'.\n"
+                "DO NOT USE for:\n"
+                "- specific named people, pets, or delivery companies → use sub_label\n"
+                "- animal species or breed names like 'blue jay', 'cardinal', "
+                "'golden retriever' → use sub_label\n"
+                "- license plate strings → use sub_label\n"
+                "- generic object queries like 'all cars today' or 'every "
+                "person' → use label alone with no semantic_query\n"
+                "When set, combine with label/time/camera/zone filters as "
+                "usual (e.g. label='person', semantic_query='riding a lawn "
+                "mower', after='2024-05-01T00:00:00Z')."
+            ),
+        }
+
+    search_objects_description = (
+        "Search the historical record of detected objects in Frigate. "
+        "Use this ONLY for questions about the PAST — e.g. 'did anyone come by today?', "
+        "'when was the last car?', 'show me detections from yesterday'. "
+        "Do NOT use this for monitoring or alerting requests about future events — "
+        "use start_camera_watch instead for those. "
+        "An 'object' in Frigate represents a tracked detection (e.g., a person, package, car).\n\n"
+        "Choose filters based on what the user is asking for:\n"
+        "- Generic class query ('show me all cars today'): set `label` only.\n"
+        "- Specific NAMED entity (known person, delivery company, animal "
+        "species/breed like 'blue jay' or 'golden retriever', license "
+        "plate): set `sub_label` only and leave `label` unset.\n"
+    )
+    if semantic_search_enabled:
+        search_objects_description += (
+            "- Physical CHARACTERISTIC, APPEARANCE, or ACTIVITY that is not a "
+            "discrete name ('person riding a lawn mower', 'someone in a red "
+            "jacket', 'person carrying a package'): set `semantic_query` with "
+            "the descriptive phrase, optionally alongside `label` for the "
+            "object class. Do NOT put descriptive phrases in sub_label."
+        )
+
     return [
         {
             "type": "function",
             "function": {
                 "name": "search_objects",
-                "description": (
-                    "Search the historical record of detected objects in Frigate. "
-                    "Use this ONLY for questions about the PAST — e.g. 'did anyone come by today?', "
-                    "'when was the last car?', 'show me detections from yesterday'. "
-                    "Do NOT use this for monitoring or alerting requests about future events — "
-                    "use start_camera_watch instead for those. "
-                    "An 'object' in Frigate represents a tracked detection (e.g., a person, package, car). "
-                    "When the user asks about a specific name (person, delivery company, animal, etc.), "
-                    "filter by sub_label only and do not set label."
-                ),
+                "description": search_objects_description,
                 "parameters": {
                     "type": "object",
-                    "properties": {
-                        "camera": {
-                            "type": "string",
-                            "description": "Camera name to filter by (optional).",
-                        },
-                        "label": {
-                            "type": "string",
-                            "description": "Object label to filter by (e.g., 'person', 'package', 'car').",
-                        },
-                        "sub_label": {
-                            "type": "string",
-                            "description": "Name of a person, delivery company, animal, etc. When filtering by a specific name, use only sub_label; do not set label.",
-                        },
-                        "after": {
-                            "type": "string",
-                            "description": "Start time in ISO 8601 format (e.g., '2024-01-01T00:00:00Z').",
-                        },
-                        "before": {
-                            "type": "string",
-                            "description": "End time in ISO 8601 format (e.g., '2024-01-01T23:59:59Z').",
-                        },
-                        "zones": {
-                            "type": "array",
-                            "items": {"type": "string"},
-                            "description": "List of zone names to filter by.",
-                        },
-                        "limit": {
-                            "type": "integer",
-                            "description": "Maximum number of objects to return (default: 25).",
-                            "default": 25,
-                        },
-                    },
+                    "properties": search_objects_properties,
                 },
                 "required": [],
             },
@@ -395,22 +458,67 @@ def get_tool_definitions() -> List[Dict[str, Any]]:
     summary="Get available tools",
     description="Returns OpenAI-compatible tool definitions for function calling.",
 )
-def get_tools() -> JSONResponse:
+def get_tools(request: Request) -> JSONResponse:
     """Get list of available tools for LLM function calling."""
-    tools = get_tool_definitions()
+    semantic_search_enabled = bool(
+        getattr(request.app.frigate_config.semantic_search, "enabled", False)
+    )
+    tools = get_tool_definitions(semantic_search_enabled=semantic_search_enabled)
     return JSONResponse(content={"tools": tools})
 
 
+def _resolve_zones(
+    zones: List[str],
+    config: FrigateConfig,
+    target_cameras: List[str],
+) -> List[str]:
+    """Map zone names to their canonical config keys, case-insensitively.
+
+    LLMs frequently echo a user's casing ("Front Yard") instead of the
+    configured key ("front_yard"). The downstream zone filter is a SQLite GLOB
+    over the JSON-encoded zones column, which is case-sensitive — so an
+    unnormalized name silently returns zero matches. Build a lookup over the
+    relevant cameras' configured zones and substitute when we find a match;
+    unknown names pass through so behavior matches what the model asked for.
+    """
+    if not zones:
+        return zones
+
+    lookup: Dict[str, str] = {}
+    for camera_id in target_cameras:
+        camera_config = config.cameras.get(camera_id)
+        if camera_config is None:
+            continue
+        for zone_name in camera_config.zones.keys():
+            lookup.setdefault(zone_name.lower(), zone_name)
+
+    return [lookup.get(z.lower(), z) for z in zones]
+
+
 async def _execute_search_objects(
+    request: Request,
     arguments: Dict[str, Any],
     allowed_cameras: List[str],
 ) -> JSONResponse:
     """
     Execute the search_objects tool.
 
-    This searches for detected objects (events) in Frigate using the same
-    logic as the events API endpoint.
+    Routes to the semantic path when the LLM supplied a `semantic_query`
+    and semantic search is enabled; otherwise delegates to the standard
+    events API logic.
     """
+    config = request.app.frigate_config
+    semantic_query = arguments.get("semantic_query")
+    if isinstance(semantic_query, str):
+        semantic_query = semantic_query.strip() or None
+    else:
+        semantic_query = None
+
+    if semantic_query and getattr(config.semantic_search, "enabled", False):
+        return await _execute_search_objects_semantic(
+            request, arguments, allowed_cameras, semantic_query
+        )
+
     # Parse after/before as server local time; convert to Unix timestamp
     after = arguments.get("after")
     before = arguments.get("before")
@@ -437,6 +545,11 @@ async def _execute_search_objects(
     # Convert zones array to comma-separated string if provided
     zones = arguments.get("zones")
     if isinstance(zones, list):
+        camera_arg = arguments.get("camera")
+        target_cameras = (
+            [camera_arg] if camera_arg and camera_arg != "all" else allowed_cameras
+        )
+        zones = _resolve_zones(zones, config, target_cameras)
         zones = ",".join(zones)
     elif zones is None:
         zones = "all"
@@ -472,6 +585,119 @@ async def _execute_search_objects(
         )
 
 
+async def _execute_search_objects_semantic(
+    request: Request,
+    arguments: Dict[str, Any],
+    allowed_cameras: List[str],
+    semantic_query: str,
+) -> JSONResponse:
+    """Search objects via fused thumbnail + description embeddings.
+
+    Runs both visual and description vec searches against `semantic_query`,
+    intersects the candidates with the structured filters (camera, label,
+    sub_label, zones, time window) the LLM supplied, and ranks the survivors
+    by fused similarity. Mirrors the candidate-then-filter pattern used by
+    find_similar_objects since sqlite-vec's IN filter is unreliable.
+    """
+    from peewee import fn
+
+    config = request.app.frigate_config
+    context = request.app.embeddings
+    if context is None:
+        logger.warning(
+            "semantic_query supplied but embeddings context is unavailable; "
+            "returning empty results."
+        )
+        return JSONResponse(content=[])
+
+    after = parse_iso_to_timestamp(arguments.get("after"))
+    before = parse_iso_to_timestamp(arguments.get("before"))
+
+    camera_arg = arguments.get("camera")
+    if camera_arg and camera_arg != "all":
+        if camera_arg not in allowed_cameras:
+            return JSONResponse(content=[])
+        cameras = [camera_arg]
+    else:
+        cameras = list(allowed_cameras) if allowed_cameras else []
+
+    if not cameras:
+        return JSONResponse(content=[])
+
+    label = arguments.get("label")
+    sub_label = arguments.get("sub_label")
+
+    zones = arguments.get("zones")
+    if isinstance(zones, list) and zones:
+        zones = _resolve_zones(zones, config, cameras)
+    else:
+        zones = None
+
+    limit = int(arguments.get("limit", 25))
+    limit = max(1, min(limit, 100))
+
+    visual_distances: Dict[str, float] = {}
+    description_distances: Dict[str, float] = {}
+    try:
+        rows = context.search_thumbnail(semantic_query)
+        visual_distances = {row[0]: row[1] for row in rows}
+    except Exception:
+        logger.exception(
+            "search_thumbnail failed for semantic_query: %s", semantic_query
+        )
+
+    try:
+        rows = context.search_description(semantic_query)
+        description_distances = {row[0]: row[1] for row in rows}
+    except Exception:
+        logger.exception(
+            "search_description failed for semantic_query: %s", semantic_query
+        )
+
+    vec_ids = set(visual_distances) | set(description_distances)
+    if not vec_ids:
+        return JSONResponse(content=[])
+
+    clauses = [Event.id.in_(list(vec_ids)), Event.camera.in_(cameras)]
+    if after is not None:
+        clauses.append(Event.start_time >= after)
+    if before is not None:
+        clauses.append(Event.start_time <= before)
+    if label:
+        clauses.append(Event.label == label)
+    if sub_label:
+        # case-insensitive match to mirror events() behavior
+        clauses.append(fn.LOWER(Event.sub_label.cast("text")) == sub_label.lower())
+    if zones:
+        zone_clauses = [Event.zones.cast("text") % f'*"{zone}"*' for zone in zones]
+        clauses.append(reduce(operator.or_, zone_clauses))
+
+    eligible = {e.id: e for e in Event.select().where(reduce(operator.and_, clauses))}
+
+    scored: List[tuple[str, float]] = []
+    for eid in eligible:
+        v_score = (
+            distance_to_score(visual_distances[eid], context.thumb_stats)
+            if eid in visual_distances
+            else None
+        )
+        d_score = (
+            distance_to_score(description_distances[eid], context.desc_stats)
+            if eid in description_distances
+            else None
+        )
+        fused = fuse_scores(v_score, d_score)
+        if fused is None:
+            continue
+        scored.append((eid, fused))
+
+    scored.sort(key=lambda pair: pair[1], reverse=True)
+    scored = scored[:limit]
+
+    results = [hydrate_event(eligible[eid], score=score) for eid, score in scored]
+    return JSONResponse(content=results)
+
+
 async def _execute_find_similar_objects(
     request: Request,
     arguments: Dict[str, Any],
@@ -528,6 +754,11 @@ async def _execute_find_similar_objects(
     sub_labels = arguments.get("sub_labels")
     zones = arguments.get("zones")
 
+    if zones:
+        zones = _resolve_zones(
+            zones, request.app.frigate_config, cameras or list(allowed_cameras)
+        )
+
     similarity_mode = arguments.get("similarity_mode", "fused")
     if similarity_mode not in ("visual", "semantic", "fused"):
         similarity_mode = "fused"
@@ -655,7 +886,7 @@ async def execute_tool(
     logger.debug(f"Executing tool: {tool_name} with arguments: {arguments}")
 
     if tool_name == "search_objects":
-        return await _execute_search_objects(arguments, allowed_cameras)
+        return await _execute_search_objects(request, arguments, allowed_cameras)
 
     if tool_name == "find_similar_objects":
         result = await _execute_find_similar_objects(
@@ -835,7 +1066,7 @@ async def _execute_tool_internal(
     This is used by the chat completion endpoint to execute tools.
     """
     if tool_name == "search_objects":
-        response = await _execute_search_objects(arguments, allowed_cameras)
+        response = await _execute_search_objects(request, arguments, allowed_cameras)
         try:
             if hasattr(response, "body"):
                 body_str = response.body.decode("utf-8")
@@ -899,6 +1130,9 @@ async def _execute_start_camera_watch(
 
     await require_camera_access(camera, request=request)
 
+    if zones:
+        zones = _resolve_zones(zones, config, [camera])
+
     genai_manager = request.app.genai_manager
     chat_client = genai_manager.chat_client
     if chat_client is None or not chat_client.supports_vision:
@@ -1245,7 +1479,9 @@ async def chat_completion(
             status_code=400,
         )
 
-    tools = get_tool_definitions()
+    config = request.app.frigate_config
+    semantic_search_enabled = bool(getattr(config.semantic_search, "enabled", False))
+    tools = get_tool_definitions(semantic_search_enabled=semantic_search_enabled)
     conversation = []
 
     current_datetime = datetime.now()
@@ -1253,7 +1489,7 @@ async def chat_completion(
     current_time_str = current_datetime.strftime("%I:%M:%S %p")
 
     cameras_info = []
-    config = request.app.frigate_config
+    has_speed_zone = False
     for camera_id in allowed_cameras:
         if camera_id not in config.cameras:
             continue
@@ -1264,6 +1500,10 @@ async def chat_completion(
             else camera_id.replace("_", " ").title()
         )
         zone_names = list(camera_config.zones.keys())
+        if not has_speed_zone:
+            has_speed_zone = any(
+                zone.distances for zone in camera_config.zones.values()
+            )
         if zone_names:
             cameras_info.append(
                 f"  - {friendly_name} (ID: {camera_id}, zones: {', '.join(zone_names)})"
@@ -1279,6 +1519,22 @@ async def chat_completion(
             + "\n\nWhen users refer to cameras by their friendly name (e.g., 'Back Deck Camera'), use the corresponding camera ID (e.g., 'back_deck_cam') in tool calls."
         )
 
+    speed_units_section = ""
+    if has_speed_zone:
+        speed_unit = (
+            "mph" if config.ui.unit_system == UnitSystemEnum.imperial else "km/h"
+        )
+        speed_units_section = f"\n\nReport object speeds to the user in {speed_unit}."
+
+    semantic_search_section = ""
+    if semantic_search_enabled:
+        semantic_search_section = (
+            "\n\nWhen routing a search_objects call, pick filters by the shape of the user's request:\n"
+            "- Generic class ('show me all cars today'): set `label` only.\n"
+            "- Specific named entity — a known person ('John'), delivery company ('Amazon'), animal species/breed ('blue jay', 'cardinal', 'golden retriever'), or license plate: set `sub_label` only and leave `label` unset.\n"
+            "- Physical characteristic, appearance, or activity that is NOT a discrete name ('find me people riding a lawn mower', 'someone in a red jacket', 'a person carrying a package'): set `semantic_query` with the descriptive phrase, optionally combined with `label` for the object class. Never put descriptive phrases in `sub_label`."
+        )
+
     system_prompt = f"""You are a helpful assistant for Frigate, a security camera NVR system. You help users answer questions about their cameras, detected objects, and events.
 
 Current server local date and time: {current_date_str} at {current_time_str}
@@ -1290,7 +1546,7 @@ When users ask about "today", "yesterday", "this week", etc., use the current da
 When searching for objects or events, use ISO 8601 format for dates (e.g., {current_date_str}T00:00:00Z for the start of today).
 Always be accurate with time calculations based on the current date provided.
 
-When a user refers to a specific object they have seen or describe with identifying details ("that green car", "the person in the red jacket", "a package left today"), prefer the find_similar_objects tool over search_objects. Use search_objects first only to locate the anchor event, then pass its id to find_similar_objects. For generic queries like "show me all cars today", keep using search_objects. If a user message begins with [attached_event:<id>], treat that event id as the anchor for any similarity or "tell me more" request in the same message and call find_similar_objects with that id.{cameras_section}"""
+When a user refers to a specific object they have seen or describe with identifying details ("that green car", "the person in the red jacket", "a package left today"), prefer the find_similar_objects tool over search_objects. Use search_objects first only to locate the anchor event, then pass its id to find_similar_objects. For generic queries like "show me all cars today", keep using search_objects. If a user message begins with [attached_event:<id>], treat that event id as the anchor for any similarity or "tell me more" request in the same message and call find_similar_objects with that id.{semantic_search_section}{cameras_section}{speed_units_section}"""
 
     conversation.append(
         {
@@ -1351,6 +1607,11 @@ When a user refers to a specific object they have seen or describe with identify
                             )
                             + b"\n"
                         )
+                    elif kind == "stats":
+                        yield (
+                            json.dumps({"type": "stats", **value}).encode("utf-8")
+                            + b"\n"
+                        )
                     elif kind == "message":
                         msg = value
                         if msg.get("finish_reason") == "error":
@@ -1581,6 +1842,7 @@ async def start_vlm_monitor(
             dispatcher=request.app.dispatcher,
             labels=body.labels,
             zones=body.zones,
+            username=request.headers.get("remote-user", ""),
         )
     except RuntimeError as e:
         logger.error("Failed to start VLM watch job: %s", e, exc_info=True)
@@ -1601,10 +1863,22 @@ async def start_vlm_monitor(
     summary="Get current VLM watch job",
     description="Returns the current (or most recently completed) VLM watch job.",
 )
-async def get_vlm_monitor() -> JSONResponse:
+async def get_vlm_monitor(request: Request) -> JSONResponse:
     job = get_vlm_watch_job()
     if job is None:
         return JSONResponse(content={"active": False}, status_code=200)
+
+    role = request.headers.get("remote-role", "viewer")
+    username = request.headers.get("remote-user", "")
+
+    # Admin and the job's creator always see the job. Other users only see it
+    # if they have access to the camera being watched; otherwise hide it.
+    if role != "admin" and username != job.username:
+        try:
+            await require_camera_access(job.camera, request=request)
+        except HTTPException:
+            return JSONResponse(content={"active": False}, status_code=200)
+
     return JSONResponse(content={"active": True, **job.to_dict()}, status_code=200)
 
 
@@ -1614,7 +1888,27 @@ async def get_vlm_monitor() -> JSONResponse:
     summary="Cancel the current VLM watch job",
     description="Cancels the running watch job if one exists.",
 )
-async def cancel_vlm_monitor() -> JSONResponse:
+async def cancel_vlm_monitor(request: Request) -> JSONResponse:
+    job = get_vlm_watch_job()
+    if job is None:
+        return JSONResponse(
+            content={"success": False, "message": "No active watch job to cancel."},
+            status_code=404,
+        )
+
+    role = request.headers.get("remote-role", "viewer")
+    username = request.headers.get("remote-user", "")
+
+    # Admin can cancel any job; other users can only cancel jobs they started.
+    if role != "admin" and username != job.username:
+        return JSONResponse(
+            content={
+                "success": False,
+                "message": "Not authorized to cancel this watch job.",
+            },
+            status_code=403,
+        )
+
     cancelled = stop_vlm_watch_job()
     if not cancelled:
         return JSONResponse(

frigate/api/debug_replay.py

@@ -10,6 +10,7 @@ from pydantic import BaseModel, Field
 
 from frigate.api.auth import require_role
 from frigate.api.defs.tags import Tags
+from frigate.jobs.debug_replay import start_debug_replay_job
 
 logger = logging.getLogger(__name__)
 
@@ -29,10 +30,17 @@ class DebugReplayStartResponse(BaseModel):
 
     success: bool
     replay_camera: str
+    job_id: str
 
 
 class DebugReplayStatusResponse(BaseModel):
-    """Response for debug replay status."""
+    """Response for debug replay status.
+
+    Returns only session-presence fields. Startup progress and error
+    details flow through the job_state WebSocket topic via the
+    debug_replay job (see frigate.jobs.debug_replay); the
+    Replay page subscribes there with useJobStatus("debug_replay").
+    """
 
     active: bool
     replay_camera: str | None = None
@@ -51,15 +59,32 @@ class DebugReplayStopResponse(BaseModel):
 @router.post(
     "/debug_replay/start",
     response_model=DebugReplayStartResponse,
+    status_code=202,
+    responses={
+        400: {"description": "Invalid camera, time range, or no recordings"},
+        409: {"description": "A replay session is already active"},
+    },
     dependencies=[Depends(require_role(["admin"]))],
     summary="Start debug replay",
-    description="Start a debug replay session from camera recordings.",
+    description="Start a debug replay session from camera recordings. Returns "
+    "immediately while clip generation runs as a background job; subscribe "
+    "to the 'debug_replay' job_state WS topic to track progress.",
 )
 async def start_debug_replay(request: Request, body: DebugReplayStartBody):
-    """Start a debug replay session."""
+    """Start a debug replay session asynchronously."""
     replay_manager = request.app.replay_manager
 
-    if replay_manager.active:
+    try:
+        job_id = await asyncio.to_thread(
+            start_debug_replay_job,
+            source_camera=body.camera,
+            start_ts=body.start_time,
+            end_ts=body.end_time,
+            frigate_config=request.app.frigate_config,
+            config_publisher=request.app.config_publisher,
+            replay_manager=replay_manager,
+        )
+    except RuntimeError:
         return JSONResponse(
             content={
                 "success": False,
@@ -67,38 +92,23 @@ async def start_debug_replay(request: Request, body: DebugReplayStartBody):
             },
             status_code=409,
         )
-
-    try:
-        replay_camera = await asyncio.to_thread(
-            replay_manager.start,
-            source_camera=body.camera,
-            start_ts=body.start_time,
-            end_ts=body.end_time,
-            frigate_config=request.app.frigate_config,
-            config_publisher=request.app.config_publisher,
-        )
     except ValueError:
-        logger.exception("Invalid parameters for debug replay start request")
+        logger.exception("Rejected debug replay start request")
         return JSONResponse(
             content={
                 "success": False,
-                "message": "Invalid debug replay request parameters",
+                "message": "Invalid debug replay parameters",
             },
             status_code=400,
         )
-    except RuntimeError:
-        logger.exception("Error while starting debug replay session")
-        return JSONResponse(
-            content={
-                "success": False,
-                "message": "An internal error occurred while starting debug replay",
-            },
-            status_code=500,
-        )
 
-    return DebugReplayStartResponse(
-        success=True,
-        replay_camera=replay_camera,
+    return JSONResponse(
+        content={
+            "success": True,
+            "replay_camera": replay_manager.replay_camera_name,
+            "job_id": job_id,
+        },
+        status_code=202,
     )
 
 
@@ -118,12 +128,16 @@ def get_debug_replay_status(request: Request):
 
     if replay_manager.active and replay_camera:
         frame_processor = request.app.detected_frames_processor
-        frame = frame_processor.get_current_frame(replay_camera)
+        frame = (
+            frame_processor.get_current_frame(replay_camera)
+            if frame_processor is not None
+            else None
+        )
 
         if frame is not None:
             frame_time = frame_processor.get_current_frame_time(replay_camera)
             camera_config = request.app.frigate_config.cameras.get(replay_camera)
-            retry_interval = 10
+            retry_interval = 10.0
 
             if camera_config is not None:
                 retry_interval = float(camera_config.ffmpeg.retry_interval or 10)

frigate/api/event.py

@@ -754,6 +754,15 @@ def events_search(
                 status_code=404,
             )
 
+        if search_event.camera not in allowed_cameras:
+            return JSONResponse(
+                content={
+                    "success": False,
+                    "message": "Event not found",
+                },
+                status_code=404,
+            )
+
         thumb_result = context.search_thumbnail(search_event)
         thumb_ids = {result[0]: result[1] for result in thumb_result}
         search_results = {

frigate/api/export.py

@@ -5,13 +5,15 @@ import logging
 import random
 import string
 import time
+import zipfile
+from collections import deque
 from pathlib import Path
-from typing import List, Optional
+from typing import Iterator, List, Optional
 
 import psutil
 from fastapi import APIRouter, Depends, Query, Request
-from fastapi.responses import JSONResponse
-from pathvalidate import sanitize_filepath
+from fastapi.responses import JSONResponse, StreamingResponse
+from pathvalidate import sanitize_filename, sanitize_filepath
 from peewee import DoesNotExist
 from playhouse.shortcuts import model_to_dict
 
@@ -361,6 +363,136 @@ def get_export_case(case_id: str):
         )
 
 
+_ZIP_STREAM_CHUNK_SIZE = 1024 * 1024  # 1 MiB
+
+
+class _StreamingZipBuffer:
+    """File-like sink for ZipFile that exposes written bytes via drain().
+
+    ZipFile writes synchronously into this buffer; the generator drains the
+    queue between writes so StreamingResponse can yield bytes without
+    materializing the whole archive in memory.
+    """
+
+    def __init__(self) -> None:
+        self._queue: deque[bytes] = deque()
+        self._offset = 0
+
+    def write(self, data: bytes) -> int:
+        if data:
+            self._queue.append(bytes(data))
+            self._offset += len(data)
+        return len(data)
+
+    def tell(self) -> int:
+        return self._offset
+
+    def flush(self) -> None:
+        pass
+
+    def drain(self) -> Iterator[bytes]:
+        while self._queue:
+            yield self._queue.popleft()
+
+
+def _unique_archive_name(export: Export, used: set[str]) -> str:
+    base = sanitize_filename(export.name) if export.name else None
+    if not base:
+        base = f"{export.camera}_{int(datetime.datetime.timestamp(export.date))}"
+
+    candidate = f"{base}.mp4"
+    counter = 1
+    while candidate in used:
+        candidate = f"{base}_{counter}.mp4"
+        counter += 1
+
+    used.add(candidate)
+    return candidate
+
+
+def _stream_case_archive(exports: List[Export]) -> Iterator[bytes]:
+    """Yield bytes of a zip archive built from the given exports' mp4 files."""
+    buffer = _StreamingZipBuffer()
+    used_names: set[str] = set()
+
+    # ZIP_STORED: mp4 is already compressed, recompressing wastes CPU for ~0% size win.
+    with zipfile.ZipFile(
+        buffer,
+        mode="w",
+        compression=zipfile.ZIP_STORED,
+        allowZip64=True,
+    ) as archive:
+        for export in exports:
+            source = Path(export.video_path)
+            if not source.exists():
+                continue
+
+            arcname = _unique_archive_name(export, used_names)
+
+            with (
+                archive.open(arcname, mode="w", force_zip64=True) as entry,
+                source.open("rb") as src,
+            ):
+                while True:
+                    chunk = src.read(_ZIP_STREAM_CHUNK_SIZE)
+                    if not chunk:
+                        break
+
+                    entry.write(chunk)
+                    yield from buffer.drain()
+
+            yield from buffer.drain()
+
+    yield from buffer.drain()
+
+
+@router.get(
+    "/cases/{case_id}/download",
+    dependencies=[Depends(allow_any_authenticated())],
+    summary="Download export case as zip",
+    description="Streams a zip archive containing every completed export's mp4 for the given case.",
+)
+def download_export_case(
+    case_id: str,
+    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+):
+    try:
+        case = ExportCase.get(ExportCase.id == case_id)
+    except DoesNotExist:
+        return JSONResponse(
+            content={"success": False, "message": "Export case not found"},
+            status_code=404,
+        )
+
+    exports = list(
+        Export.select()
+        .where(
+            Export.export_case == case_id,
+            ~Export.in_progress,
+            Export.camera << allowed_cameras,
+        )
+        .order_by(Export.date.asc())
+    )
+
+    if not exports:
+        return JSONResponse(
+            content={"success": False, "message": "No exports available to download."},
+            status_code=404,
+        )
+
+    archive_base = sanitize_filename(case.name) if case.name else ""
+    if not archive_base:
+        archive_base = case_id
+
+    return StreamingResponse(
+        _stream_case_archive(exports),
+        media_type="application/zip",
+        headers={
+            "Content-Disposition": f'attachment; filename="{archive_base}.zip"',
+        },
+    )
+
+
 @router.patch(
     "/cases/{case_id}",
     response_model=GenericResponse,

frigate/api/media.py

@@ -174,12 +174,10 @@ async def latest_frame(
     }
     quality_params = get_image_quality_params(extension.value, params.quality)
 
-    if camera_name in request.app.frigate_config.cameras:
+    camera_config = request.app.frigate_config.cameras.get(camera_name)
+    if camera_config is not None:
         frame = frame_processor.get_current_frame(camera_name, draw_options)
-        retry_interval = float(
-            request.app.frigate_config.cameras.get(camera_name).ffmpeg.retry_interval
-            or 10
-        )
+        retry_interval = float(camera_config.ffmpeg.retry_interval or 10)
 
         is_offline = False
         if frame is None or datetime.now().timestamp() > (
@@ -1368,12 +1366,17 @@ def preview_gif(
         file_start = f"preview_{camera_name}-"
         start_file = f"{file_start}{start_ts}.{PREVIEW_FRAME_TYPE}"
         end_file = f"{file_start}{end_ts}.{PREVIEW_FRAME_TYPE}"
+
+        camera_files = [
+            entry.name
+            for entry in os.scandir(preview_dir)
+            if entry.name.startswith(file_start)
+        ]
+        camera_files.sort()
+
         selected_previews = []
 
-        for file in sorted(os.listdir(preview_dir)):
-            if not file.startswith(file_start):
-                continue
-
+        for file in camera_files:
             if file < start_file:
                 continue
 
@@ -1550,12 +1553,17 @@ def preview_mp4(
         file_start = f"preview_{camera_name}-"
         start_file = f"{file_start}{start_ts}.{PREVIEW_FRAME_TYPE}"
         end_file = f"{file_start}{end_ts}.{PREVIEW_FRAME_TYPE}"
+
+        camera_files = [
+            entry.name
+            for entry in os.scandir(preview_dir)
+            if entry.name.startswith(file_start)
+        ]
+        camera_files.sort()
+
         selected_previews = []
 
-        for file in sorted(os.listdir(preview_dir)):
-            if not file.startswith(file_start):
-                continue
-
+        for file in camera_files:
             if file < start_file:
                 continue
 

frigate/api/media_auth.py

@@ -0,0 +1,291 @@
+"""URI-aware authorization for nginx-served static media.
+
+The `/auth` endpoint (used as nginx `auth_request` target) calls into this
+module to classify the requested URI from the `X-Original-URL` header and, for
+camera-scoped resources, decide whether the current role may access them.
+
+Without this, `auth_request` only verifies the JWT — every authenticated user
+could read clips, recordings, and exports for *any* camera, bypassing the
+per-camera authorization the regular API enforces via `require_camera_access`.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+from enum import Enum
+from typing import Optional
+from urllib.parse import unquote, urlparse
+
+from peewee import DoesNotExist
+
+from frigate.config import FrigateConfig
+from frigate.const import EXPORT_DIR
+from frigate.models import Export, User
+
+logger = logging.getLogger(__name__)
+
+
+class MediaAuthResolution(str, Enum):
+    """Classification of an `X-Original-URL` path for media-auth purposes."""
+
+    CAMERA = "camera"
+    ADMIN_ONLY = "admin_only"
+    LISTING_MULTI_CAMERA = "listing_multi_camera"
+    LISTING_NEUTRAL = "listing_neutral"
+    # Under a recognized media root (/clips, /recordings, /exports) but
+    # unclassifiable (unknown subtree, no matching DB row, DB error).
+    # Restricted users are denied; admins/full-access roles are allowed
+    # (nginx will likely return 404 if the file genuinely doesn't exist).
+    UNRESOLVED_MEDIA = "unresolved_media"
+    # Not a media URI at all (e.g. /api/events, /login).
+    UNKNOWN = "unknown"
+
+
+def extract_path(original_url: Optional[str]) -> Optional[str]:
+    """Return the decoded path component of nginx's `X-Original-URL` header.
+
+    nginx forwards the *raw* request URI (with `..` segments intact) via
+    `$request_uri`. nginx normalizes the path before serving the file, so a
+    request like `/recordings/.../allowed_cam/../forbidden_cam/file.mp4`
+    would (1) parse as the allowed camera in our auth check, (2) be served
+    as the forbidden camera by nginx. To close the bypass we reject any URI
+    whose path contains `.` or `..` segments outright.
+    """
+    if not original_url:
+        return None
+
+    parsed = urlparse(original_url)
+    raw_path = parsed.path or original_url
+    decoded = unquote(raw_path)
+    if not decoded:
+        return None
+
+    if not decoded.startswith("/"):
+        decoded = "/" + decoded
+
+    segments = decoded.split("/")
+    if ".." in segments or "." in segments:
+        return None
+
+    return decoded
+
+
+def resolve_media_uri(
+    uri: str, frigate_config: Optional[FrigateConfig] = None
+) -> tuple[MediaAuthResolution, Optional[str]]:
+    """Classify a URI and return the owning camera if applicable.
+
+    `frigate_config` is used to disambiguate clip/review filenames whose
+    camera name contains hyphens by matching against the longest configured
+    camera-name prefix.
+    """
+    if not uri:
+        return MediaAuthResolution.UNKNOWN, None
+
+    parts = [p for p in uri.split("/") if p]
+    if not parts:
+        return MediaAuthResolution.UNKNOWN, None
+
+    root = parts[0]
+    if root == "recordings":
+        return _resolve_recording(parts)
+    if root == "clips":
+        return _resolve_clip(parts, frigate_config)
+    if root == "exports":
+        return _resolve_export(parts)
+
+    return MediaAuthResolution.UNKNOWN, None
+
+
+def _resolve_recording(
+    parts: list[str],
+) -> tuple[MediaAuthResolution, Optional[str]]:
+    # /recordings                          → neutral
+    # /recordings/{date}                   → neutral
+    # /recordings/{date}/{hour}            → multi-camera listing
+    # /recordings/{date}/{hour}/{cam}/...  → camera
+    if len(parts) <= 2:
+        return MediaAuthResolution.LISTING_NEUTRAL, None
+    if len(parts) == 3:
+        return MediaAuthResolution.LISTING_MULTI_CAMERA, None
+    return MediaAuthResolution.CAMERA, parts[3]
+
+
+def _resolve_clip(
+    parts: list[str], frigate_config: Optional[FrigateConfig]
+) -> tuple[MediaAuthResolution, Optional[str]]:
+    # /clips                                          → multi-camera listing
+    # /clips/thumbs/{cam}/...                         → camera
+    # /clips/previews/{cam}/...                       → camera
+    # /clips/review/thumb-{cam}-{review_id}.webp      → camera (parsed)
+    # /clips/faces/...                                → admin-only
+    # /clips/genai-requests/...                       → admin-only
+    # /clips/preview_restart_cache/...                → admin-only
+    # /clips/{model}/train|dataset/...                → admin-only
+    # /clips/{cam}-{event_id}[-clean].{ext}           → camera (parsed)
+    # other /clips/{subdir}/...                       → unresolved (deny restricted)
+    if len(parts) == 1:
+        return MediaAuthResolution.LISTING_MULTI_CAMERA, None
+
+    second = parts[1]
+
+    if second in ("thumbs", "previews"):
+        if len(parts) == 2:
+            return MediaAuthResolution.LISTING_MULTI_CAMERA, None
+        return MediaAuthResolution.CAMERA, parts[2]
+
+    if second == "review":
+        if len(parts) == 2:
+            return MediaAuthResolution.LISTING_MULTI_CAMERA, None
+        camera = _camera_from_thumb_filename(parts[2], frigate_config)
+        if camera:
+            return MediaAuthResolution.CAMERA, camera
+        return MediaAuthResolution.UNRESOLVED_MEDIA, None
+
+    if second in ("faces", "genai-requests", "preview_restart_cache"):
+        return MediaAuthResolution.ADMIN_ONLY, None
+
+    if len(parts) >= 3 and parts[2] in ("train", "dataset"):
+        return MediaAuthResolution.ADMIN_ONLY, None
+
+    if len(parts) == 2:
+        camera = _camera_from_clip_filename(second, frigate_config)
+        if camera:
+            return MediaAuthResolution.CAMERA, camera
+        return MediaAuthResolution.UNRESOLVED_MEDIA, None
+
+    return MediaAuthResolution.UNRESOLVED_MEDIA, None
+
+
+def _longest_prefix_camera(
+    stem: str, frigate_config: Optional[FrigateConfig]
+) -> Optional[str]:
+    if frigate_config is None:
+        return None
+    for cam in sorted(frigate_config.cameras.keys(), key=len, reverse=True):
+        if stem.startswith(cam + "-"):
+            return cam
+    return None
+
+
+def _camera_from_clip_filename(
+    filename: str, frigate_config: Optional[FrigateConfig]
+) -> Optional[str]:
+    """Match a flat clip filename `{camera}-{event_id}[-clean].{ext}` against
+    configured camera names. Longest-prefix wins so camera names containing
+    hyphens (e.g. `front-door`) resolve correctly.
+    """
+    dot = filename.rfind(".")
+    stem = filename[:dot] if dot > 0 else filename
+    return _longest_prefix_camera(stem, frigate_config)
+
+
+def _camera_from_thumb_filename(
+    filename: str, frigate_config: Optional[FrigateConfig]
+) -> Optional[str]:
+    """Match a review thumbnail filename `thumb-{camera}-{review_id}.webp`."""
+    if not filename.startswith("thumb-"):
+        return None
+    dot = filename.rfind(".")
+    stem = filename[len("thumb-") : dot] if dot > 0 else filename[len("thumb-") :]
+    return _longest_prefix_camera(stem, frigate_config)
+
+
+def _resolve_export(
+    parts: list[str],
+) -> tuple[MediaAuthResolution, Optional[str]]:
+    # /exports                  → multi-camera listing
+    # /exports/{filename}.mp4   → camera (DB lookup by exact path)
+    if len(parts) == 1:
+        return MediaAuthResolution.LISTING_MULTI_CAMERA, None
+    if len(parts) != 2:
+        return MediaAuthResolution.UNRESOLVED_MEDIA, None
+
+    filename = parts[1]
+    full_path = os.path.join(EXPORT_DIR, filename)
+    try:
+        export = Export.get(Export.video_path == full_path)
+        return MediaAuthResolution.CAMERA, export.camera
+    except DoesNotExist:
+        return MediaAuthResolution.UNRESOLVED_MEDIA, None
+    except Exception as e:
+        logger.warning("Export DB lookup failed for %s: %s", filename, e)
+        return MediaAuthResolution.UNRESOLVED_MEDIA, None
+
+
+def check_camera_access(role: str, camera: str, frigate_config: FrigateConfig) -> bool:
+    """Return True iff `role` may access `camera`.
+
+    Mirrors the gating logic in `require_camera_access`: admin and any role
+    without a non-empty allow-list bypass the check.
+    """
+    if role == "admin":
+        return True
+
+    roles_dict = frigate_config.auth.roles
+    if not roles_dict.get(role):
+        return True
+
+    all_camera_names = set(frigate_config.cameras.keys())
+    allowed = User.get_allowed_cameras(role, roles_dict, all_camera_names)
+    return camera in allowed
+
+
+def is_role_restricted(role: str, frigate_config: FrigateConfig) -> bool:
+    """True if `role` has a non-empty allow-list (i.e. not full-access)."""
+    if role == "admin":
+        return False
+    return bool(frigate_config.auth.roles.get(role))
+
+
+def deny_response_for_media_uri(
+    original_url: Optional[str], role: Optional[str], frigate_config: FrigateConfig
+) -> Optional[int]:
+    """Decide whether the current role should be blocked from `original_url`.
+
+    Returns an HTTP status code (403) when access should be denied, or `None`
+    when the request is allowed.
+    """
+    if not original_url:
+        return None
+
+    path = extract_path(original_url)
+
+    # `extract_path` returns None for URIs containing `.` or `..` segments.
+    # For media-root URIs that's a traversal attempt — deny outright. For
+    # non-media URIs, pass through (nginx / the backend handle them).
+    if path is None:
+        raw = urlparse(original_url).path or original_url
+        decoded = unquote(raw)
+        first = decoded.lstrip("/").split("/", 1)[0] if decoded else ""
+        if first in ("clips", "recordings", "exports"):
+            return 403
+        return None
+
+    resolution, camera = resolve_media_uri(path, frigate_config)
+    if resolution == MediaAuthResolution.UNKNOWN:
+        return None
+
+    if not role or role == "admin":
+        return None
+
+    if not is_role_restricted(role, frigate_config):
+        return None
+
+    if resolution == MediaAuthResolution.LISTING_NEUTRAL:
+        return None
+
+    if resolution in (
+        MediaAuthResolution.LISTING_MULTI_CAMERA,
+        MediaAuthResolution.ADMIN_ONLY,
+        MediaAuthResolution.UNRESOLVED_MEDIA,
+    ):
+        return 403
+
+    if resolution == MediaAuthResolution.CAMERA:
+        if camera and check_camera_access(role, camera, frigate_config):
+            return None
+        return 403
+
+    return 403

frigate/api/preview.py

@@ -148,12 +148,17 @@ def get_preview_frames_from_cache(camera_name: str, start_ts: float, end_ts: flo
     file_start = f"preview_{camera_name}-"
     start_file = f"{file_start}{start_ts}.{PREVIEW_FRAME_TYPE}"
     end_file = f"{file_start}{end_ts}.{PREVIEW_FRAME_TYPE}"
+
+    camera_files = [
+        entry.name
+        for entry in os.scandir(preview_dir)
+        if entry.name.startswith(file_start)
+    ]
+    camera_files.sort()
+
     selected_previews = []
 
-    for file in sorted(os.listdir(preview_dir)):
-        if not file.startswith(file_start):
-            continue
-
+    for file in camera_files:
         if file < start_file:
             continue
 

frigate/api/record.py

@@ -35,7 +35,7 @@ logger = logging.getLogger(__name__)
 router = APIRouter(tags=[Tags.recordings])
 
 
-@router.get("/recordings/storage", dependencies=[Depends(allow_any_authenticated())])
+@router.get("/recordings/storage", dependencies=[Depends(require_role(["admin"]))])
 def get_recordings_storage_usage(request: Request):
     recording_stats = request.app.stats_emitter.get_latest_stats()["service"][
         "storage"

frigate/app.py

@@ -144,7 +144,7 @@ class FrigateApp:
         for d in dirs:
             if not os.path.exists(d) and not os.path.islink(d):
                 logger.info(f"Creating directory: {d}")
-                os.makedirs(d)
+                os.makedirs(d, exist_ok=True)
             else:
                 logger.debug(f"Skipping directory: {d}")
 
@@ -189,17 +189,6 @@ class FrigateApp:
             except PermissionError:
                 logger.error("Unable to write to /config to save DB state")
 
-        def cleanup_timeline_db(db: SqliteExtDatabase) -> None:
-            db.execute_sql(
-                "DELETE FROM timeline WHERE source_id NOT IN (SELECT id FROM event);"
-            )
-
-            try:
-                with open(f"{CONFIG_DIR}/.timeline", "w") as f:
-                    f.write(str(datetime.datetime.now().timestamp()))
-            except PermissionError:
-                logger.error("Unable to write to /config to save DB state")
-
         # Migrate DB schema
         migrate_db = SqliteExtDatabase(self.config.database.path)
 
@@ -216,11 +205,6 @@ class FrigateApp:
 
         router.run()
 
-        # this is a temporary check to clean up user DB from beta
-        # will be removed before final release
-        if not os.path.exists(f"{CONFIG_DIR}/.timeline"):
-            cleanup_timeline_db(migrate_db)
-
         # check if vacuum needs to be run
         if os.path.exists(f"{CONFIG_DIR}/.vacuum"):
             with open(f"{CONFIG_DIR}/.vacuum") as f:
@@ -444,18 +428,11 @@ class FrigateApp:
         self.camera_maintainer.start()
 
     def start_audio_processor(self) -> None:
-        audio_cameras = [
-            c
-            for c in self.config.cameras.values()
-            if c.enabled and c.audio.enabled_in_config
-        ]
-
-        if audio_cameras:
-            self.audio_process = AudioProcessor(
-                self.config, audio_cameras, self.camera_metrics, self.stop_event
-            )
-            self.audio_process.start()
-            self.processes["audio_detector"] = self.audio_process.pid or 0
+        self.audio_process = AudioProcessor(
+            self.config, self.camera_metrics, self.stop_event
+        )
+        self.audio_process.start()
+        self.processes["audio_detector"] = self.audio_process.pid or 0
 
     def start_timeline_processor(self) -> None:
         self.timeline_processor = TimelineProcessor(

frigate/comms/webpush.py

@@ -429,7 +429,10 @@ class WebPushClient(Communicator):
             else:
                 title = base_title
 
-            message = payload["after"]["data"]["metadata"]["shortSummary"]
+            if payload["after"]["data"]["metadata"].get("shortSummary"):
+                message = payload["after"]["data"]["metadata"]["shortSummary"]
+            else:
+                message = f"Detected on {camera_name}"
         else:
             zone_names = payload["after"]["data"]["zones"]
             formatted_zone_names = []
@@ -549,6 +552,14 @@ class WebPushClient(Communicator):
         logger.debug(f"Sending camera monitoring push notification for {camera_name}")
 
         for user in self.web_pushers:
+            if not self._user_has_camera_access(user, camera):
+                logger.debug(
+                    "Skipping notification for user %s - no access to camera %s",
+                    user,
+                    camera,
+                )
+                continue
+
             self.send_push_notification(
                 user=user,
                 payload=payload,

frigate/comms/ws.py

@@ -17,9 +17,90 @@ from ws4py.websocket import WebSocket as WebSocket_
 
 from frigate.comms.base_communicator import Communicator
 from frigate.config import FrigateConfig
+from frigate.const import (
+    CLEAR_ONGOING_REVIEW_SEGMENTS,
+    EXPIRE_AUDIO_ACTIVITY,
+    INSERT_MANY_RECORDINGS,
+    INSERT_PREVIEW,
+    NOTIFICATION_TEST,
+    REQUEST_REGION_GRID,
+    UPDATE_AUDIO_ACTIVITY,
+    UPDATE_AUDIO_TRANSCRIPTION_STATE,
+    UPDATE_BIRDSEYE_LAYOUT,
+    UPDATE_CAMERA_ACTIVITY,
+    UPDATE_EMBEDDINGS_REINDEX_PROGRESS,
+    UPDATE_EVENT_DESCRIPTION,
+    UPDATE_MODEL_STATE,
+    UPDATE_REVIEW_DESCRIPTION,
+    UPSERT_REVIEW_SEGMENT,
+)
 
 logger = logging.getLogger(__name__)
 
+# Internal IPC topics — NEVER allowed from WebSocket, regardless of role
+_WS_BLOCKED_TOPICS = frozenset(
+    {
+        INSERT_MANY_RECORDINGS,
+        INSERT_PREVIEW,
+        REQUEST_REGION_GRID,
+        UPSERT_REVIEW_SEGMENT,
+        CLEAR_ONGOING_REVIEW_SEGMENTS,
+        UPDATE_CAMERA_ACTIVITY,
+        UPDATE_AUDIO_ACTIVITY,
+        EXPIRE_AUDIO_ACTIVITY,
+        UPDATE_EVENT_DESCRIPTION,
+        UPDATE_REVIEW_DESCRIPTION,
+        UPDATE_MODEL_STATE,
+        UPDATE_EMBEDDINGS_REINDEX_PROGRESS,
+        UPDATE_BIRDSEYE_LAYOUT,
+        UPDATE_AUDIO_TRANSCRIPTION_STATE,
+        NOTIFICATION_TEST,
+    }
+)
+
+# Read-only topics any authenticated user (including viewer) can send
+_WS_VIEWER_TOPICS = frozenset(
+    {
+        "onConnect",
+        "modelState",
+        "audioTranscriptionState",
+        "birdseyeLayout",
+        "embeddingsReindexProgress",
+    }
+)
+
+
+def _check_ws_authorization(
+    topic: str,
+    role_header: str | None,
+    separator: str,
+) -> bool:
+    """Check if a WebSocket message is authorized.
+
+    Args:
+        topic: The message topic.
+        role_header: The HTTP_REMOTE_ROLE header value, or None.
+        separator: The role separator character from proxy config.
+
+    Returns:
+        True if authorized, False if blocked.
+    """
+    # Block IPC-only topics unconditionally
+    if topic in _WS_BLOCKED_TOPICS:
+        return False
+
+    # No role header: default to viewer (fail-closed)
+    if role_header is None:
+        return topic in _WS_VIEWER_TOPICS
+
+    # Check if any role is admin
+    roles = [r.strip() for r in role_header.split(separator)]
+    if "admin" in roles:
+        return True
+
+    # Non-admin: only viewer topics allowed
+    return topic in _WS_VIEWER_TOPICS
+
 
 class WebSocket(WebSocket_):  # type: ignore[misc]
     def unhandled_error(self, error: Any) -> None:
@@ -49,6 +130,7 @@ class WebSocketClient(Communicator):
 
         class _WebSocketHandler(WebSocket):
             receiver = self._dispatcher
+            role_separator = self.config.proxy.separator or ","
 
             def received_message(self, message: WebSocket.received_message) -> None:  # type: ignore[name-defined]
                 try:
@@ -63,11 +145,25 @@ class WebSocketClient(Communicator):
                     )
                     return
 
-                logger.debug(
-                    f"Publishing mqtt message from websockets at {json_message['topic']}."
+                topic = json_message["topic"]
+
+                # Authorization check (skip when environ is None — direct internal connection)
+                role_header = (
+                    self.environ.get("HTTP_REMOTE_ROLE") if self.environ else None
                 )
+                if self.environ is not None and not _check_ws_authorization(
+                    topic, role_header, self.role_separator
+                ):
+                    logger.warning(
+                        "Blocked unauthorized WebSocket message: topic=%s, role=%s",
+                        topic,
+                        role_header,
+                    )
+                    return
+
+                logger.debug(f"Publishing mqtt message from websockets at {topic}.")
                 self.receiver(
-                    json_message["topic"],
+                    topic,
                     json_message["payload"],
                 )
 

frigate/config/camera/camera.py

@@ -76,7 +76,7 @@ class CameraConfig(FrigateBaseModel):
     # Options with global fallback
     audio: AudioConfig = Field(
         default_factory=AudioConfig,
-        title="Audio events",
+        title="Audio detection",
         description="Settings for audio-based event detection for this camera.",
     )
     audio_transcription: CameraAudioTranscriptionConfig = Field(

frigate/config/camera/genai.py

@@ -41,8 +41,7 @@ class GenAIConfig(FrigateBaseModel):
         title="Model",
         description="The model to use from the provider for generating descriptions or summaries.",
     )
-    provider: GenAIProviderEnum | None = Field(
-        default=None,
+    provider: GenAIProviderEnum = Field(
         title="Provider",
         description="The GenAI provider to use (for example: ollama, gemini, openai).",
     )

frigate/config/camera/updater.py

@@ -20,6 +20,7 @@ class CameraConfigUpdateEnum(str, Enum):
     ffmpeg = "ffmpeg"
     live = "live"
     motion = "motion"  # includes motion and motion masks
+    mqtt = "mqtt"
     notifications = "notifications"
     objects = "objects"
     object_genai = "object_genai"
@@ -33,6 +34,7 @@ class CameraConfigUpdateEnum(str, Enum):
     lpr = "lpr"
     snapshots = "snapshots"
     timestamp_style = "timestamp_style"
+    ui = "ui"
     zones = "zones"
 
 
@@ -119,7 +121,10 @@ class CameraConfigUpdateSubscriber:
         elif update_type == CameraConfigUpdateEnum.objects:
             config.objects = updated_config
         elif update_type == CameraConfigUpdateEnum.record:
+            old_enabled_in_config = config.record.enabled_in_config
             config.record = updated_config
+            if old_enabled_in_config != updated_config.enabled_in_config:
+                config.recreate_ffmpeg_cmds()
         elif update_type == CameraConfigUpdateEnum.review:
             config.review = updated_config
         elif update_type == CameraConfigUpdateEnum.review_genai:

frigate/config/classification.py

@@ -26,6 +26,11 @@ class EnrichmentsDeviceEnum(str, Enum):
     CPU = "CPU"
 
 
+class ModelSizeEnum(str, Enum):
+    small = "small"
+    large = "large"
+
+
 class TriggerType(str, Enum):
     THUMBNAIL = "thumbnail"
     DESCRIPTION = "description"
@@ -53,13 +58,13 @@ class AudioTranscriptionConfig(FrigateBaseModel):
         title="Transcription language",
         description="Language code used for transcription/translation (for example 'en' for English). See https://whisper-api.com/docs/languages/ for supported language codes.",
     )
-    device: Optional[EnrichmentsDeviceEnum] = Field(
+    device: EnrichmentsDeviceEnum = Field(
         default=EnrichmentsDeviceEnum.CPU,
         title="Transcription device",
         description="Device key (CPU/GPU) to run the transcription model on. Only NVIDIA CUDA GPUs are currently supported for transcription.",
     )
-    model_size: str = Field(
-        default="small",
+    model_size: ModelSizeEnum = Field(
+        default=ModelSizeEnum.small,
         title="Model size",
         description="Model size to use for offline audio event transcription.",
     )
@@ -189,8 +194,8 @@ class SemanticSearchConfig(FrigateBaseModel):
                 return v
         return v
 
-    model_size: str = Field(
-        default="small",
+    model_size: ModelSizeEnum = Field(
+        default=ModelSizeEnum.small,
         title="Model size",
         description="Select model size; 'small' runs on CPU and 'large' typically requires GPU.",
     )
@@ -253,8 +258,8 @@ class FaceRecognitionConfig(FrigateBaseModel):
         title="Enable face recognition",
         description="Enable or disable face recognition for all cameras; can be overridden per-camera.",
     )
-    model_size: str = Field(
-        default="small",
+    model_size: ModelSizeEnum = Field(
+        default=ModelSizeEnum.small,
         title="Model size",
         description="Model size to use for face embeddings (small/large); larger may require GPU.",
     )
@@ -335,8 +340,8 @@ class LicensePlateRecognitionConfig(FrigateBaseModel):
         title="Enable LPR",
         description="Enable or disable license plate recognition for all cameras; can be overridden per-camera.",
     )
-    model_size: str = Field(
-        default="small",
+    model_size: ModelSizeEnum = Field(
+        default=ModelSizeEnum.small,
         title="Model size",
         description="Model size used for text detection/recognition. Most users should use 'small'.",
     )

frigate/config/config.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import io
 import json
 import logging
 import os
@@ -25,7 +26,6 @@ from frigate.plus import PlusApi
 from frigate.util.builtin import (
     deep_merge,
     get_ffmpeg_arg_list,
-    load_labels,
 )
 from frigate.util.config import (
     CURRENT_CONFIG_VERSION,
@@ -80,17 +80,40 @@ logger = logging.getLogger(__name__)
 
 yaml = YAML()
 
+DEFAULT_DETECTORS = {
+    "ov": {
+        "type": "openvino",
+        "device": "CPU",
+    }
+}
+DEFAULT_MODEL = {
+    "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",
+}
+DEFAULT_DETECT_DIMENSIONS = {"width": 1280, "height": 720}
+
+
+def _render_default_yaml(data: dict) -> str:
+    buf = io.StringIO()
+    _yaml_writer = YAML()
+    _yaml_writer.indent(mapping=2, sequence=4, offset=2)
+    _yaml_writer.dump(data, buf)
+    return buf.getvalue()
+
+
 DEFAULT_CONFIG = f"""
 mqtt:
   enabled: False
 
+{_render_default_yaml({"detectors": DEFAULT_DETECTORS, "model": DEFAULT_MODEL})}
 cameras: {{}}  # No cameras defined, UI wizard should be used
 version: {CURRENT_CONFIG_VERSION}
 """
 
-DEFAULT_DETECTORS = {"cpu": {"type": "cpu"}}
-DEFAULT_DETECT_DIMENSIONS = {"width": 1280, "height": 720}
-
 # stream info handler
 stream_info_retriever = StreamInfoRetriever()
 
@@ -453,7 +476,7 @@ class FrigateConfig(FrigateBaseModel):
     cameras: Dict[str, CameraConfig] = Field(title="Cameras", description="Cameras")
     audio: AudioConfig = Field(
         default_factory=AudioConfig,
-        title="Audio events",
+        title="Audio detection",
         description="Settings for audio-based event detection for all cameras; can be overridden per-camera.",
     )
     birdseye: BirdseyeConfig = Field(
@@ -614,17 +637,12 @@ class FrigateConfig(FrigateBaseModel):
         if self.ffmpeg.hwaccel_args == "auto":
             self.ffmpeg.hwaccel_args = auto_detect_hwaccel()
 
-        # Populate global audio filters for all audio labels
-        all_audio_labels = {
-            label
-            for label in load_labels("/audio-labelmap.txt", prefill=521).values()
-            if label
-        }
-
+        # Populate global audio filters from listen. Existing user-defined
+        # entries for labels not in listen are preserved but unused at runtime.
         if self.audio.filters is None:
             self.audio.filters = {}
 
-        for key in sorted(all_audio_labels - self.audio.filters.keys()):
+        for key in sorted(set(self.audio.listen) - self.audio.filters.keys()):
             self.audio.filters[key] = AudioFilterConfig()
 
         self.audio.filters = dict(sorted(self.audio.filters.items()))
@@ -679,6 +697,9 @@ class FrigateConfig(FrigateBaseModel):
                     model_config["path"] = "/cpu_model.tflite"
                 elif detector_config.type == "edgetpu":
                     model_config["path"] = "/edgetpu_model.tflite"
+                elif detector_config.type == "openvino":
+                    for default_key, default_value in DEFAULT_MODEL.items():
+                        model_config.setdefault(default_key, default_value)
 
             model = ModelConfig.model_validate(model_config)
             model.check_and_load_plus_model(self.plus_api, detector_config.type)
@@ -813,7 +834,9 @@ class FrigateConfig(FrigateBaseModel):
             if camera_config.audio.filters is None:
                 camera_config.audio.filters = {}
 
-            for key in sorted(all_audio_labels - camera_config.audio.filters.keys()):
+            for key in sorted(
+                set(camera_config.audio.listen) - camera_config.audio.filters.keys()
+            ):
                 camera_config.audio.filters[key] = AudioFilterConfig()
 
             camera_config.audio.filters = dict(
@@ -835,7 +858,9 @@ class FrigateConfig(FrigateBaseModel):
                     if mask_config:
                         coords = mask_config.coordinates
                         relative_coords = get_relative_coordinates(
-                            coords, camera_config.frame_shape
+                            coords,
+                            camera_config.frame_shape,
+                            camera_name=camera_config.name,
                         )
                         # Create a new ObjectMaskConfig with raw_coordinates set
                         processed_global_masks[mask_id] = ObjectMaskConfig(

frigate/config/telemetry.py

@@ -25,8 +25,8 @@ class StatsConfig(FrigateBaseModel):
     )
     intel_gpu_device: Optional[str] = Field(
         default=None,
-        title="SR-IOV device",
-        description="Device identifier used when treating Intel GPUs as SR-IOV to fix GPU stats.",
+        title="Intel GPU device",
+        description="PCI bus address or DRM device path (e.g. /dev/dri/card1) used to pin Intel GPU stats to a specific device when multiple are present.",
     )
 
 

frigate/const.py

@@ -15,7 +15,7 @@ TRIGGER_DIR = f"{CLIPS_DIR}/triggers"
 BIRDSEYE_PIPE = "/tmp/cache/birdseye"
 CACHE_DIR = "/tmp/cache"
 REPLAY_CAMERA_PREFIX = "_replay_"
-REPLAY_DIR = os.path.join(CACHE_DIR, "replay")
+REPLAY_DIR = os.path.join(CLIPS_DIR, "replay")
 PLUS_ENV_VAR = "PLUS_API_KEY"
 PLUS_API_HOST = "https://api.frigate.video"
 

frigate/data_processing/common/face/model.py

@@ -133,6 +133,61 @@ class FaceRecognizer(ABC):
             return 0.0
 
 
+def build_class_mean(
+    embs: list[np.ndarray],
+    trim: float = 0.15,
+    outlier_threshold: float = 0.30,
+    min_keep_frac: float = 0.7,
+    max_iters: int = 3,
+) -> np.ndarray:
+    """Build a class-mean embedding with two-layer outlier protection.
+
+    Layer 1 (iterative, vector-wise): drop whole embeddings whose cosine
+    similarity to the current class mean is below ``outlier_threshold``.
+    Catches mislabeled or corrupted training samples (wrong face in the
+    folder, full-frame screenshots, extreme crops) that per-dimension
+    trimming cannot detect.
+
+    Layer 2 (per-dimension): ``scipy.stats.trim_mean`` on the retained set
+    to smooth per-component noise (lighting, expression, alignment jitter).
+
+    Collections with fewer than 5 images bypass outlier rejection — too few
+    samples to establish a reliable class center.
+    """
+    arr = np.stack(embs, axis=0)
+
+    if len(arr) < 5:
+        return np.asarray(stats.trim_mean(arr, trim, axis=0))
+
+    keep = np.ones(len(arr), dtype=bool)
+    floor = max(5, int(np.ceil(min_keep_frac * len(arr))))
+
+    for _ in range(max_iters):
+        mean = stats.trim_mean(arr[keep], trim, axis=0)
+        m_norm = mean / (np.linalg.norm(mean) + 1e-9)
+        e_norms = arr / (np.linalg.norm(arr, axis=1, keepdims=True) + 1e-9)
+        cos = e_norms @ m_norm
+        new_keep = cos >= outlier_threshold
+
+        if new_keep.sum() < floor:
+            top = np.argsort(-cos)[:floor]
+            new_keep = np.zeros(len(arr), dtype=bool)
+            new_keep[top] = True
+
+        if np.array_equal(new_keep, keep):
+            break
+        keep = new_keep
+
+    dropped = int((~keep).sum())
+
+    if dropped:
+        logger.debug(
+            f"Vector-wise outlier filter dropped {dropped}/{len(arr)} embeddings"
+        )
+
+    return np.asarray(stats.trim_mean(arr[keep], trim, axis=0))
+
+
 def similarity_to_confidence(
     cosine_similarity: float,
     median: float = 0.3,
@@ -229,7 +284,7 @@ class FaceNetRecognizer(FaceRecognizer):
 
         for name, embs in face_embeddings_map.items():
             if embs:
-                self.mean_embs[name] = stats.trim_mean(embs, 0.15)
+                self.mean_embs[name] = build_class_mean(embs)
 
         logger.debug("Finished building ArcFace model")
 
@@ -340,7 +395,7 @@ class ArcFaceRecognizer(FaceRecognizer):
 
         for name, embs in face_embeddings_map.items():
             if embs:
-                self.mean_embs[name] = stats.trim_mean(embs, 0.15)
+                self.mean_embs[name] = build_class_mean(embs)
 
         logger.debug("Finished building ArcFace model")
 

frigate/data_processing/common/license_plate/mixin.py

@@ -1073,10 +1073,6 @@ class LicensePlateProcessingMixin:
                     top_score = score
                     top_box = bbox
 
-                if score > top_score:
-                    top_score = score
-                    top_box = bbox
-
         # Return the top scoring bounding box if found
         if top_box is not None:
             # expand box by 5% to help with OCR
@@ -1092,9 +1088,6 @@ class LicensePlateProcessingMixin:
                 ]
             ).clip(0, [input.shape[1], input.shape[0]] * 2)
 
-            logger.debug(
-                f"{camera}: Found license plate. Bounding box: {expanded_box.astype(int)}"
-            )
             return tuple(int(x) for x in expanded_box)  # type: ignore[return-value]
         else:
             return None  # No detection above the threshold
@@ -1360,8 +1353,8 @@ class LicensePlateProcessingMixin:
                 )
 
                 # check that license plate is valid
-                # double the value because we've doubled the size of the car
-                if license_plate_area < self.config.cameras[camera].lpr.min_area * 2:
+                # quadruple the value because we've doubled both dimensions of the car
+                if license_plate_area < self.config.cameras[camera].lpr.min_area * 4:
                     logger.debug(f"{camera}: License plate is less than min_area")
                     return
 
@@ -1465,6 +1458,7 @@ class LicensePlateProcessingMixin:
                     license_plate_frame,
                 )
 
+        logger.debug(f"{camera}: Found license plate. Bounding box: {list(plate_box)}")
         logger.debug(f"{camera}: Running plate recognition for id: {id}.")
 
         # run detection, returns results sorted by confidence, best first

frigate/data_processing/post/object_descriptions.py

@@ -269,7 +269,9 @@ class ObjectDescriptionProcessor(PostProcessorApi):
 
         if event.has_snapshot and camera_config.objects.genai.use_snapshot:
             snapshot_image = self._read_and_crop_snapshot(event)
+
             if not snapshot_image:
+                self.cleanup_event(event_id)
                 return
 
         num_thumbnails = len(self.tracked_events.get(event_id, []))

frigate/data_processing/post/review_descriptions.py

@@ -39,6 +39,8 @@ logger = logging.getLogger(__name__)
 
 RECORDING_BUFFER_EXTENSION_PERCENT = 0.10
 MIN_RECORDING_DURATION = 10
+MAX_IMAGE_TOKENS = 24000
+MAX_FRAMES_PER_SECOND = 1
 
 
 class ReviewDescriptionProcessor(PostProcessorApi):
@@ -60,14 +62,22 @@ class ReviewDescriptionProcessor(PostProcessorApi):
     def calculate_frame_count(
         self,
         camera: str,
+        duration: float,
         image_source: ImageSourceEnum = ImageSourceEnum.preview,
         height: int = 480,
     ) -> int:
-        """Calculate optimal number of frames based on context size, image source, and resolution.
+        """Calculate optimal number of frames based on event duration, context size,
+        image source, and resolution.
 
-        Token usage varies by resolution: larger images (ultra-wide aspect ratios) use more tokens.
-        Estimates ~1 token per 1250 pixels. Targets 98% context utilization with safety margin.
-        Capped at 20 frames.
+        Per-image token cost is asked of the GenAI provider so providers that know
+        their model's true cost (e.g. llama.cpp can probe the loaded mmproj) can
+        diverge from the default ~1-token-per-1250-pixels heuristic. The frame
+        budget is bounded by:
+          - remaining context window after prompt + response reservations
+          - a fixed MAX_IMAGE_TOKENS ceiling
+          - MAX_FRAMES_PER_SECOND x duration, to avoid drowning short events in
+            near-duplicate frames where the model latches onto the redundant middle
+            and skips the start/end action
         """
         client = self.genai_manager.description_client
 
@@ -105,14 +115,15 @@ class ReviewDescriptionProcessor(PostProcessorApi):
                 width = target_width
                 height = int(target_width / aspect_ratio)
 
-        pixels_per_image = width * height
-        tokens_per_image = pixels_per_image / 1250
+        tokens_per_image = client.estimate_image_tokens(width, height)
         prompt_tokens = 3800
         response_tokens = 300
-        available_tokens = context_size - prompt_tokens - response_tokens
-        max_frames = int(available_tokens / tokens_per_image)
-
-        return min(max(max_frames, 3), 20)
+        context_budget = context_size - prompt_tokens - response_tokens
+        image_token_budget = min(context_budget, MAX_IMAGE_TOKENS)
+        max_frames_by_tokens = int(image_token_budget / tokens_per_image)
+        max_frames_by_duration = int(duration * MAX_FRAMES_PER_SECOND)
+        max_frames = min(max_frames_by_tokens, max_frames_by_duration)
+        return max(max_frames, 3)
 
     def process_data(
         self, data: dict[str, Any], data_type: PostProcessDataEnum
@@ -355,12 +366,17 @@ class ReviewDescriptionProcessor(PostProcessorApi):
         file_start = f"preview_{camera}-"
         start_file = f"{file_start}{start_time}.webp"
         end_file = f"{file_start}{end_time}.webp"
+
+        camera_files = [
+            entry.name
+            for entry in os.scandir(preview_dir)
+            if entry.name.startswith(file_start)
+        ]
+        camera_files.sort()
+
         all_frames: list[str] = []
 
-        for file in sorted(os.listdir(preview_dir)):
-            if not file.startswith(file_start):
-                continue
-
+        for file in camera_files:
             if file < start_file:
                 if len(all_frames):
                     all_frames[0] = os.path.join(preview_dir, file)
@@ -376,7 +392,9 @@ class ReviewDescriptionProcessor(PostProcessorApi):
             all_frames.append(os.path.join(preview_dir, file))
 
         frame_count = len(all_frames)
-        desired_frame_count = self.calculate_frame_count(camera)
+        desired_frame_count = self.calculate_frame_count(
+            camera, duration=end_time - start_time
+        )
 
         if frame_count <= desired_frame_count:
             return all_frames
@@ -400,7 +418,7 @@ class ReviewDescriptionProcessor(PostProcessorApi):
         """Get frames from recordings at specified timestamps."""
         duration = end_time - start_time
         desired_frame_count = self.calculate_frame_count(
-            camera, ImageSourceEnum.recordings, height
+            camera, duration, ImageSourceEnum.recordings, height
         )
 
         # Calculate evenly spaced timestamps throughout the duration

frigate/data_processing/post/types.py

@@ -1,21 +1,37 @@
-from pydantic import BaseModel, ConfigDict, Field
+from typing import Annotated
+
+from pydantic import BaseModel, ConfigDict, Field, StringConstraints
+
+ObservationItem = Annotated[str, StringConstraints(min_length=20, max_length=200)]
 
 
 class ReviewMetadata(BaseModel):
     model_config = ConfigDict(extra="ignore", protected_namespaces=())
 
-    title: str = Field(
-        description="A short title characterizing what took place and where, under 10 words."
+    observations: list[ObservationItem] = Field(
+        ...,
+        min_length=3,
+        max_length=8,
+        description="Enumerate the significant observations across all frames, in chronological order.",
     )
     scene: str = Field(
-        description="A chronological narrative of what happens from start to finish."
+        min_length=150,
+        max_length=600,
+        description="A chronological narrative of what happens from start to finish, drawing directly from the items in observations.",
+    )
+    title: str = Field(
+        max_length=80,
+        description="Title for the activity.",
     )
     shortSummary: str = Field(
-        description="A brief 2-sentence summary of the scene, suitable for notifications."
+        min_length=70,
+        max_length=140,
+        description="A brief summary for the activity.",
     )
     confidence: float = Field(
         ge=0.0,
-        description="Confidence in the analysis, from 0 to 1.",
+        le=1.0,
+        description="Confidence in the analysis as a decimal between 0.0 and 1.0, where 0.0 means no confidence and 1.0 means complete confidence. Express ONLY as a decimal.",
     )
     potential_threat_level: int = Field(
         ge=0,

frigate/data_processing/real_time/face.py

@@ -229,9 +229,10 @@ class FaceRealTimeProcessor(RealTimeProcessorApi):
                 logger.debug(f"No person box available for {id}")
                 return
 
-            rgb = cv2.cvtColor(frame, cv2.COLOR_YUV2RGB_I420)
+            # YuNet (cv2.FaceDetectorYN) is trained on BGR
+            bgr = cv2.cvtColor(frame, cv2.COLOR_YUV2BGR_I420)
             left, top, right, bottom = person_box
-            person = rgb[top:bottom, left:right]
+            person = bgr[top:bottom, left:right]
             face_box = self.__detect_face(person, self.face_config.detection_threshold)
 
             if not face_box:
@@ -250,11 +251,6 @@ class FaceRealTimeProcessor(RealTimeProcessorApi):
                 )
                 return
 
-            try:
-                face_frame = cv2.cvtColor(face_frame, cv2.COLOR_RGB2BGR)
-            except Exception as e:
-                logger.debug(f"Failed to convert face frame color for {id}: {e}")
-                return
         else:
             # don't run for object without attributes
             if not obj_data.get("current_attributes"):

frigate/debug_replay.py

@@ -1,9 +1,13 @@
-"""Debug replay camera management for replaying recordings with detection overlays."""
+"""Debug replay camera management for replaying recordings with detection overlays.
+
+The startup work (ffmpeg concat + camera config publish) lives in
+frigate.jobs.debug_replay. This module owns only session presence
+(active), session metadata, and post-session cleanup.
+"""
 
 import logging
 import os
 import shutil
-import subprocess as sp
 import threading
 
 from ruamel.yaml import YAML
@@ -21,7 +25,7 @@ from frigate.const import (
     REPLAY_DIR,
     THUMB_DIR,
 )
-from frigate.models import Recordings
+from frigate.jobs.debug_replay import cancel_debug_replay_job, wait_for_runner
 from frigate.util.camera_cleanup import cleanup_camera_db, cleanup_camera_files
 from frigate.util.config import find_config_file
 
@@ -29,7 +33,14 @@ logger = logging.getLogger(__name__)
 
 
 class DebugReplayManager:
-    """Manages a single debug replay session."""
+    """Owns the lifecycle pointers for a single debug replay session.
+
+    A session exists from the moment mark_starting is called (synchronously,
+    inside the API handler) until clear_session runs (on success cleanup,
+    failure, or stop). The active property is the source of truth that the
+    status bar consumes — broader than the startup job, which only covers the
+    preparing_clip / starting_camera window.
+    """
 
     def __init__(self) -> None:
         self._lock = threading.Lock()
@@ -41,144 +52,66 @@ class DebugReplayManager:
 
     @property
     def active(self) -> bool:
-        """Whether a replay session is currently active."""
+        """True from mark_starting until clear_session."""
         return self.replay_camera_name is not None
 
-    def start(
+    def mark_starting(
         self,
         source_camera: str,
+        replay_camera_name: str,
         start_ts: float,
         end_ts: float,
-        frigate_config: FrigateConfig,
-        config_publisher: CameraConfigUpdatePublisher,
-    ) -> str:
-        """Start a debug replay session.
+    ) -> None:
+        """Synchronously claim the session before the job runner starts.
 
-        Args:
-            source_camera: Name of the source camera to replay
-            start_ts: Start timestamp
-            end_ts: End timestamp
-            frigate_config: Current Frigate configuration
-            config_publisher: Publisher for camera config updates
-
-        Returns:
-            The replay camera name
-
-        Raises:
-            ValueError: If a session is already active or parameters are invalid
-            RuntimeError: If clip generation fails
+        Called inside the API handler so the status bar sees active=True
+        immediately, before the worker thread does any ffmpeg work.
         """
         with self._lock:
-            return self._start_locked(
-                source_camera, start_ts, end_ts, frigate_config, config_publisher
-            )
+            self.replay_camera_name = replay_camera_name
+            self.source_camera = source_camera
+            self.start_ts = start_ts
+            self.end_ts = end_ts
+            self.clip_path = None
 
-    def _start_locked(
+    def mark_session_ready(self, clip_path: str) -> None:
+        """Record the on-disk clip path after the camera has been published."""
+        with self._lock:
+            self.clip_path = clip_path
+
+    def clear_session(self) -> None:
+        """Reset session pointers without publishing camera removal.
+
+        Used by the job runner on failure paths. stop() does the camera
+        teardown plus this clear in one step.
+        """
+        with self._lock:
+            self._clear_locked()
+
+    def _clear_locked(self) -> None:
+        self.replay_camera_name = None
+        self.source_camera = None
+        self.clip_path = None
+        self.start_ts = None
+        self.end_ts = None
+
+    def publish_camera(
         self,
         source_camera: str,
-        start_ts: float,
-        end_ts: float,
+        replay_name: str,
+        clip_path: str,
         frigate_config: FrigateConfig,
         config_publisher: CameraConfigUpdatePublisher,
-    ) -> str:
-        if self.active:
-            raise ValueError("A replay session is already active")
+    ) -> None:
+        """Build the in-memory replay camera config and publish the add event.
 
-        if source_camera not in frigate_config.cameras:
-            raise ValueError(f"Camera '{source_camera}' not found")
-
-        if end_ts <= start_ts:
-            raise ValueError("End time must be after start time")
-
-        # Query recordings for the source camera in the time range
-        recordings = (
-            Recordings.select(
-                Recordings.path,
-                Recordings.start_time,
-                Recordings.end_time,
-            )
-            .where(
-                Recordings.start_time.between(start_ts, end_ts)
-                | Recordings.end_time.between(start_ts, end_ts)
-                | ((start_ts > Recordings.start_time) & (end_ts < Recordings.end_time))
-            )
-            .where(Recordings.camera == source_camera)
-            .order_by(Recordings.start_time.asc())
-        )
-
-        if not recordings.count():
-            raise ValueError(
-                f"No recordings found for camera '{source_camera}' in the specified time range"
-            )
-
-        # Create replay directory
-        os.makedirs(REPLAY_DIR, exist_ok=True)
-
-        # Generate replay camera name
-        replay_name = f"{REPLAY_CAMERA_PREFIX}{source_camera}"
-
-        # Build concat file for ffmpeg
-        concat_file = os.path.join(REPLAY_DIR, f"{replay_name}_concat.txt")
-        clip_path = os.path.join(REPLAY_DIR, f"{replay_name}.mp4")
-
-        with open(concat_file, "w") as f:
-            for recording in recordings:
-                f.write(f"file '{recording.path}'\n")
-
-        # Concatenate recordings into a single clip with -c copy (fast)
-        ffmpeg_cmd = [
-            frigate_config.ffmpeg.ffmpeg_path,
-            "-hide_banner",
-            "-y",
-            "-f",
-            "concat",
-            "-safe",
-            "0",
-            "-i",
-            concat_file,
-            "-c",
-            "copy",
-            "-movflags",
-            "+faststart",
-            clip_path,
-        ]
-
-        logger.info(
-            "Generating replay clip for %s (%.1f - %.1f)",
-            source_camera,
-            start_ts,
-            end_ts,
-        )
-
-        try:
-            result = sp.run(
-                ffmpeg_cmd,
-                capture_output=True,
-                text=True,
-                timeout=120,
-            )
-            if result.returncode != 0:
-                logger.error("FFmpeg error: %s", result.stderr)
-                raise RuntimeError(
-                    f"Failed to generate replay clip: {result.stderr[-500:]}"
-                )
-        except sp.TimeoutExpired:
-            raise RuntimeError("Clip generation timed out")
-        finally:
-            # Clean up concat file
-            if os.path.exists(concat_file):
-                os.remove(concat_file)
-
-        if not os.path.exists(clip_path):
-            raise RuntimeError("Clip file was not created")
-
-        # Build camera config dict for the replay camera
+        Called by the job runner during the starting_camera phase.
+        """
         source_config = frigate_config.cameras[source_camera]
         camera_dict = self._build_camera_config_dict(
             source_config, replay_name, clip_path
         )
 
-        # Build an in-memory config with the replay camera added
         config_file = find_config_file()
         yaml_parser = YAML()
         with open(config_file, "r") as f:
@@ -191,75 +124,48 @@ class DebugReplayManager:
         try:
             new_config = FrigateConfig.parse_object(config_data)
         except Exception as e:
-            raise RuntimeError(f"Failed to validate replay camera config: {e}")
-
-        # Update the running config
+            raise RuntimeError(f"Failed to validate replay camera config: {e}") from e
         frigate_config.cameras[replay_name] = new_config.cameras[replay_name]
 
-        # Publish the add event
         config_publisher.publish_update(
             CameraConfigUpdateTopic(CameraConfigUpdateEnum.add, replay_name),
             new_config.cameras[replay_name],
         )
 
-        # Store session state
-        self.replay_camera_name = replay_name
-        self.source_camera = source_camera
-        self.clip_path = clip_path
-        self.start_ts = start_ts
-        self.end_ts = end_ts
-
-        logger.info("Debug replay started: %s -> %s", source_camera, replay_name)
-        return replay_name
-
     def stop(
         self,
         frigate_config: FrigateConfig,
         config_publisher: CameraConfigUpdatePublisher,
     ) -> None:
-        """Stop the active replay session and clean up all artifacts.
+        """Cancel any in-flight startup job and tear down the active session.
 
-        Args:
-            frigate_config: Current Frigate configuration
-            config_publisher: Publisher for camera config updates
+        Safe to call when no session is active (no-op with a warning).
         """
+        cancel_debug_replay_job()
+        wait_for_runner(timeout=2.0)
+
         with self._lock:
-            self._stop_locked(frigate_config, config_publisher)
+            if not self.active:
+                logger.warning("No active replay session to stop")
+                return
 
-    def _stop_locked(
-        self,
-        frigate_config: FrigateConfig,
-        config_publisher: CameraConfigUpdatePublisher,
-    ) -> None:
-        if not self.active:
-            logger.warning("No active replay session to stop")
-            return
+            replay_name = self.replay_camera_name
 
-        replay_name = self.replay_camera_name
+            # Only publish remove if the camera was actually added to the live
+            # config (i.e. the runner reached the starting_camera phase).
+            if replay_name is not None and replay_name in frigate_config.cameras:
+                config_publisher.publish_update(
+                    CameraConfigUpdateTopic(CameraConfigUpdateEnum.remove, replay_name),
+                    frigate_config.cameras[replay_name],
+                )
 
-        # Publish remove event so subscribers stop and remove from their config
-        if replay_name in frigate_config.cameras:
-            config_publisher.publish_update(
-                CameraConfigUpdateTopic(CameraConfigUpdateEnum.remove, replay_name),
-                frigate_config.cameras[replay_name],
-            )
-            # Do NOT pop here — let subscribers handle removal from the shared
-            # config dict when they process the ZMQ message to avoid race conditions
+            if replay_name is not None:
+                self._cleanup_db(replay_name)
+                self._cleanup_files(replay_name)
 
-        # Defensive DB cleanup
-        self._cleanup_db(replay_name)
+            self._clear_locked()
 
-        # Remove filesystem artifacts
-        self._cleanup_files(replay_name)
-
-        # Reset state
-        self.replay_camera_name = None
-        self.source_camera = None
-        self.clip_path = None
-        self.start_ts = None
-        self.end_ts = None
-
-        logger.info("Debug replay stopped and cleaned up: %s", replay_name)
+            logger.info("Debug replay stopped and cleaned up: %s", replay_name)
 
     def _build_camera_config_dict(
         self,
@@ -267,16 +173,7 @@ class DebugReplayManager:
         replay_name: str,
         clip_path: str,
     ) -> dict:
-        """Build a camera config dictionary for the replay camera.
-
-        Args:
-            source_config: Source camera's CameraConfig
-            replay_name: Name for the replay camera
-            clip_path: Path to the replay clip file
-
-        Returns:
-            Camera config as a dictionary
-        """
+        """Build a camera config dictionary for the replay camera."""
         # Extract detect config (exclude computed fields)
         detect_dict = source_config.detect.model_dump(
             exclude={"min_initialized", "max_disappeared", "enabled_in_config"}
@@ -311,7 +208,6 @@ class DebugReplayManager:
             zone_dump = zone_config.model_dump(
                 exclude={"contour", "color"}, exclude_defaults=True
             )
-            # Always include required fields
             zone_dump.setdefault("coordinates", zone_config.coordinates)
             zones_dict[zone_name] = zone_dump
 

frigate/detectors/detection_runners.py

@@ -79,7 +79,11 @@ def is_openvino_gpu_npu_available() -> bool:
     available_devices = get_openvino_available_devices()
     # Check for GPU, NPU, or other acceleration devices (excluding CPU)
     acceleration_devices = ["GPU", "MYRIAD", "NPU", "GNA", "HDDL"]
-    return any(device in available_devices for device in acceleration_devices)
+    return any(
+        avail_dev == accel_dev or avail_dev.startswith(accel_dev + ".")
+        for avail_dev in available_devices
+        for accel_dev in acceleration_devices
+    )
 
 
 class BaseModelRunner(ABC):
@@ -132,7 +136,6 @@ class ONNXModelRunner(BaseModelRunner):
         return model_type in [
             EnrichmentModelTypeEnum.paddleocr.value,
             EnrichmentModelTypeEnum.jina_v2.value,
-            EnrichmentModelTypeEnum.arcface.value,
             ModelTypeEnum.rfdetr.value,
             ModelTypeEnum.dfine.value,
         ]

frigate/detectors/plugins/openvino.py

@@ -52,6 +52,12 @@ class OvDetector(DetectionApi):
         self.h = detector_config.model.height
         self.w = detector_config.model.width
 
+        logger.info(
+            "Loading OpenVINO model %s on device %s",
+            detector_config.model.path,
+            detector_config.device,
+        )
+
         self.runner = OpenVINOModelRunner(
             model_path=detector_config.model.path,
             device=detector_config.device,

frigate/embeddings/__init__.py

@@ -4,6 +4,7 @@ import base64
 import json
 import logging
 import os
+import sys
 import threading
 from json.decoder import JSONDecodeError
 from multiprocessing.synchronize import Event as MpEvent
@@ -52,6 +53,14 @@ class EmbeddingProcess(FrigateProcess):
             self.stop_event,
         )
         maintainer.start()
+        maintainer.join()
+
+        # If the maintainer thread exited but no shutdown was requested, it
+        # crashed. Surface as a non-zero exit so the watchdog restarts us
+        # instead of treating the silent thread death as a clean shutdown.
+        if not self.stop_event.is_set():
+            logger.error("Embeddings maintainer thread exited unexpectedly")
+            sys.exit(1)
 
 
 class EmbeddingsContext:

frigate/embeddings/maintainer.py

@@ -60,7 +60,11 @@ from frigate.data_processing.real_time.license_plate import (
 )
 from frigate.data_processing.types import DataProcessorMetrics, PostProcessDataEnum
 from frigate.db.sqlitevecq import SqliteVecQueueDatabase
-from frigate.events.types import EventTypeEnum, RegenerateDescriptionEnum
+from frigate.events.types import (
+    EventStateEnum,
+    EventTypeEnum,
+    RegenerateDescriptionEnum,
+)
 from frigate.genai import GenAIClientManager
 from frigate.models import Event, Recordings, ReviewSegment, Trigger
 from frigate.types import TrackedObjectUpdateTypesEnum
@@ -310,6 +314,10 @@ class EmbeddingMaintainer(threading.Thread):
             self._handle_custom_classification_update(topic, payload)
             return
 
+        if topic == "config/genai":
+            self.config.genai = payload
+            self.genai_manager.update_config(self.config)
+
         # Broadcast to all processors — each decides if the topic is relevant
         for processor in self.realtime_processors:
             processor.update_config(topic, payload)
@@ -431,7 +439,7 @@ class EmbeddingMaintainer(threading.Thread):
         if update is None:
             return
 
-        source_type, _, camera, frame_name, data = update
+        source_type, event_type, camera, frame_name, data = update
 
         logger.debug(
             f"Received update - source_type: {source_type}, camera: {camera}, data label: {data.get('label') if data else 'None'}"
@@ -481,6 +489,12 @@ class EmbeddingMaintainer(threading.Thread):
 
         for processor in self.post_processors:
             if isinstance(processor, ObjectDescriptionProcessor):
+                # skip end events — _process_finalized handles them via event_end_subscriber.
+                # processing them here can re-create tracked_events entries after cleanup
+                # when the event_subscriber queue is backlogged behind event_end_subscriber.
+                if event_type == EventStateEnum.end:
+                    continue
+
                 processor.process_data(
                     {
                         "camera": camera,
@@ -513,10 +527,16 @@ class EmbeddingMaintainer(threading.Thread):
                 try:
                     event: Event = Event.get(Event.id == event_id)
                 except DoesNotExist:
+                    for processor in self.post_processors:
+                        if isinstance(processor, ObjectDescriptionProcessor):
+                            processor.cleanup_event(event_id)
                     continue
 
                 # Skip the event if not an object
                 if event.data.get("type") != "object":
+                    for processor in self.post_processors:
+                        if isinstance(processor, ObjectDescriptionProcessor):
+                            processor.cleanup_event(event_id)
                     continue
 
                 # Extract valid thumbnail

frigate/events/audio.py

@@ -84,7 +84,6 @@ class AudioProcessor(FrigateProcess):
     def __init__(
         self,
         config: FrigateConfig,
-        cameras: list[CameraConfig],
         camera_metrics: DictProxy,
         stop_event: MpEvent,
     ):
@@ -93,12 +92,11 @@ class AudioProcessor(FrigateProcess):
         )
 
         self.camera_metrics = camera_metrics
-        self.cameras = cameras
         self.config = config
 
     def run(self) -> None:
         self.pre_run_setup(self.config.logger)
-        audio_threads: list[AudioEventMaintainer] = []
+        audio_threads: dict[str, AudioEventMaintainer] = {}
 
         threading.current_thread().name = "process:audio_manager"
 
@@ -112,32 +110,56 @@ class AudioProcessor(FrigateProcess):
         else:
             self.transcription_model_runner = None
 
-        if len(self.cameras) == 0:
-            return
+        config_subscriber = CameraConfigUpdateSubscriber(
+            self.config,
+            self.config.cameras,
+            [
+                CameraConfigUpdateEnum.add,
+                CameraConfigUpdateEnum.audio,
+                CameraConfigUpdateEnum.ffmpeg,
+            ],
+        )
 
-        for camera in self.cameras:
-            audio_thread = AudioEventMaintainer(
+        def spawn_if_needed(camera: CameraConfig) -> None:
+            name = camera.name
+            if name is None or name in audio_threads:
+                return
+            if not camera.enabled or not camera.audio.enabled:
+                return
+            # ffmpeg update may not have arrived yet; wait for next poll
+            if not any("audio" in i.roles for i in camera.ffmpeg.inputs):
+                return
+            thread = AudioEventMaintainer(
                 camera,
                 self.config,
                 self.camera_metrics,
                 self.transcription_model_runner,
                 self.stop_event,  # type: ignore[arg-type]
             )
-            audio_threads.append(audio_thread)
-            audio_thread.start()
+            audio_threads[name] = thread
+            thread.start()
+            self.logger.info(f"Audio maintainer started for {name}")
+
+        for camera in self.config.cameras.values():
+            spawn_if_needed(camera)
 
         self.logger.info(f"Audio processor started (pid: {self.pid})")
 
-        while not self.stop_event.wait():
-            pass
+        # poll for newly added cameras or cameras flipped to audio.enabled at runtime
+        while not self.stop_event.wait(timeout=1.0):
+            config_subscriber.check_for_updates()
+            for camera in self.config.cameras.values():
+                spawn_if_needed(camera)
 
-        for thread in audio_threads:
+        config_subscriber.stop()
+
+        for thread in audio_threads.values():
             thread.join(1)
             if thread.is_alive():
                 self.logger.info(f"Waiting for thread {thread.name:s} to exit")
                 thread.join(10)
 
-        for thread in audio_threads:
+        for thread in audio_threads.values():
             if thread.is_alive():
                 self.logger.warning(f"Thread {thread.name} is still alive")
 
@@ -205,6 +227,7 @@ class AudioEventMaintainer(threading.Thread):
             self.transcription_thread.start()
 
         self.was_enabled = camera.enabled
+        self.was_audio_enabled = camera.audio.enabled
 
     def detect_audio(self, audio: np.ndarray) -> None:
         if not self.camera_config.audio.enabled or self.stop_event.is_set():
@@ -363,6 +386,17 @@ class AudioEventMaintainer(threading.Thread):
                 time.sleep(0.1)
                 continue
 
+            audio_enabled = self.camera_config.audio.enabled
+            if audio_enabled != self.was_audio_enabled:
+                if not audio_enabled:
+                    self.logger.debug(
+                        f"Disabling audio detections for {self.camera_config.name}, ending events"
+                    )
+                    self.requestor.send_data(
+                        EXPIRE_AUDIO_ACTIVITY, self.camera_config.name
+                    )
+                self.was_audio_enabled = audio_enabled
+
             self.read_audio()
 
         if self.audio_listener:

frigate/genai/__init__.py

@@ -2,6 +2,7 @@
 
 import datetime
 import importlib
+import json
 import logging
 import os
 import re
@@ -9,6 +10,7 @@ from typing import Any, Callable, Optional
 
 import numpy as np
 from playhouse.shortcuts import model_to_dict
+from pydantic import ValidationError
 
 from frigate.config import CameraConfig, GenAIConfig, GenAIProviderEnum
 from frigate.const import CLIPS_DIR
@@ -106,10 +108,11 @@ When forming your description:
 ## Response Field Guidelines
 
 Respond with a JSON object matching the provided schema. Field-specific guidance:
+- `observations`: Include the very start of the activity — for example, a vehicle entering the frame or pulling into the driveway — even if it lasts only a few frames and the rest of the clip is dominated by a longer activity. Include each arrival, departure, object handled, and notable change in position or state. Each item is a single concrete fact written as a complete sentence.
 - `scene`: Describe how the sequence begins, then the progression of events — all significant movements and actions in order. For example, if a vehicle arrives and then a person exits, describe both sequentially. For named subjects (those with a `←` separator in "Objects in Scene"), always use their name — do not replace them with generic terms. For unnamed objects (e.g., "person", "car"), refer to them naturally with articles (e.g., "a person", "the car"). Your description should align with and support the threat level you assign.
-- `title`: Characterize **what took place and where** — interpret the overall purpose or outcome, do not simply compress the scene description into fewer words. Include the relevant location (zone, area, or entry point). For named subjects, always use their name. For unnamed objects, refer to them naturally with articles. No editorial qualifiers like "routine" or "suspicious."
+- `title`: Name the primary activity across the observations, together with the location. An activity is what is being done with objects, tools, or surfaces; locomotion through the scene qualifies as the activity only when no other interaction is observed. For named subjects, always use their name. For unnamed objects, refer to them naturally with articles.
+- `shortSummary`: Briefly summarize the primary activity across the observations.
 - `potential_threat_level`: Must be consistent with your scene description and the activity patterns above.
-{get_concern_prompt()}
 
 ## Sequence Details
 
@@ -151,9 +154,6 @@ Each line represents a detection state, not necessarily unique individuals. The
             if "other_concerns" in schema.get("required", []):
                 schema["required"].remove("other_concerns")
 
-        # OpenAI strict mode requires additionalProperties: false on all objects
-        schema["additionalProperties"] = False
-
         response_format = {
             "type": "json_schema",
             "json_schema": {
@@ -181,7 +181,36 @@ Each line represents a detection state, not necessarily unique individuals. The
 
             try:
                 metadata = ReviewMetadata.model_validate_json(clean_json)
+            except ValidationError as ve:
+                # Constraint violations (length, item count, ranges) are logged
+                # at debug and the response is kept anyway — a slightly
+                # off-spec answer is still usable, and dropping the whole
+                # response loses the narrative content the model produced.
+                for err in ve.errors():
+                    loc = ".".join(str(p) for p in err["loc"]) or "<root>"
+                    logger.debug(
+                        "Review metadata soft validation: %s — %s (input: %r)",
+                        loc,
+                        err["msg"],
+                        err.get("input"),
+                    )
+                try:
+                    raw = json.loads(clean_json)
+                except json.JSONDecodeError as je:
+                    logger.error("Failed to parse review description JSON: %s", je)
+                    return None
+                # observations and confidence are required on the model; fill an empty default
+                # if the response omitted it so attribute access stays safe.
+                raw.setdefault("observations", [])
+                raw.setdefault("confidence", 0.0)
+                metadata = ReviewMetadata.model_construct(**raw)
+            except Exception as e:
+                logger.error(
+                    f"Failed to parse review description as the response did not match expected format. {e}"
+                )
+                return None
 
+            try:
                 # Normalize confidence if model returned a percentage (e.g. 85 instead of 0.85)
                 if metadata.confidence > 1.0:
                     metadata.confidence = min(metadata.confidence / 100.0, 1.0)
@@ -194,10 +223,7 @@ Each line represents a detection state, not necessarily unique individuals. The
                 metadata.time = review_data["start"]
                 return metadata
             except Exception as e:
-                # rarely LLMs can fail to follow directions on output format
-                logger.warning(
-                    f"Failed to parse review description as the response did not match expected format. {e}"
-                )
+                logger.error(f"Failed to post-process review metadata: {e}")
                 return None
         else:
             logger.debug(
@@ -344,6 +370,14 @@ Guidelines:
         """Get the context window size for this provider in tokens."""
         return 4096
 
+    def estimate_image_tokens(self, width: int, height: int) -> float:
+        """Estimate prompt tokens consumed by a single image of the given dimensions.
+
+        Default heuristic: ~1 token per 1250 pixels. Providers that can measure or
+        know their model's exact image-token cost should override.
+        """
+        return (width * height) / 1250
+
     def embed(
         self,
         texts: list[str] | None = None,

frigate/genai/azure-openai.py

@@ -10,6 +10,7 @@ from openai import AzureOpenAI
 
 from frigate.config import GenAIProviderEnum
 from frigate.genai import GenAIClient, register_genai_provider
+from frigate.genai.openai import _stats_from_openai_usage
 
 logger = logging.getLogger(__name__)
 
@@ -210,6 +211,7 @@ class OpenAIClient(GenAIClient):
                 "messages": messages,
                 "timeout": self.timeout,
                 "stream": True,
+                "stream_options": {"include_usage": True},
             }
 
             if tools:
@@ -221,10 +223,15 @@ class OpenAIClient(GenAIClient):
             content_parts: list[str] = []
             tool_calls_by_index: dict[int, dict[str, Any]] = {}
             finish_reason = "stop"
+            usage_stats: Optional[dict[str, Any]] = None
 
             stream = self.provider.chat.completions.create(**request_params)  # type: ignore[call-overload]
 
             for chunk in stream:
+                chunk_usage = getattr(chunk, "usage", None)
+                if chunk_usage is not None:
+                    usage_stats = _stats_from_openai_usage(chunk_usage)
+
                 if not chunk or not chunk.choices:
                     continue
 
@@ -284,6 +291,9 @@ class OpenAIClient(GenAIClient):
                     )
                 finish_reason = "tool_calls"
 
+            if usage_stats is not None:
+                yield ("stats", usage_stats)
+
             yield (
                 "message",
                 {

frigate/genai/gemini.py

@@ -14,6 +14,20 @@ from frigate.genai import GenAIClient, register_genai_provider
 logger = logging.getLogger(__name__)
 
 
+def _stats_from_gemini_usage(usage: Any) -> Optional[dict[str, Any]]:
+    """Build a stats dict from a Gemini usage_metadata object."""
+    prompt_tokens = getattr(usage, "prompt_token_count", None)
+    completion_tokens = getattr(usage, "candidates_token_count", None)
+    if prompt_tokens is None and completion_tokens is None:
+        return None
+    stats: dict[str, Any] = {}
+    if isinstance(prompt_tokens, int):
+        stats["prompt_tokens"] = prompt_tokens
+    if isinstance(completion_tokens, int):
+        stats["completion_tokens"] = completion_tokens
+    return stats or None
+
+
 @register_genai_provider(GenAIProviderEnum.gemini)
 class GeminiClient(GenAIClient):
     """Generative AI client for Frigate using Gemini."""
@@ -136,22 +150,44 @@ class GeminiClient(GenAIClient):
                             )
                         )
                 elif role == "assistant":
-                    gemini_messages.append(
-                        types.Content(
-                            role="model", parts=[types.Part.from_text(text=content)]
-                        )
-                    )
+                    parts: list[types.Part] = []
+                    if content:
+                        parts.append(types.Part.from_text(text=content))
+                    for tc in msg.get("tool_calls") or []:
+                        func = tc.get("function") or {}
+                        tc_name = func.get("name") or ""
+                        tc_args: Any = func.get("arguments")
+                        if isinstance(tc_args, str):
+                            try:
+                                tc_args = json.loads(tc_args)
+                            except (json.JSONDecodeError, TypeError):
+                                tc_args = {}
+                        if not isinstance(tc_args, dict):
+                            tc_args = {}
+                        if tc_name:
+                            parts.append(
+                                types.Part.from_function_call(
+                                    name=tc_name, args=tc_args
+                                )
+                            )
+                    if not parts:
+                        parts.append(types.Part.from_text(text=" "))
+                    gemini_messages.append(types.Content(role="model", parts=parts))
                 elif role == "tool":
                     # Handle tool response
-                    function_response = {
-                        "name": msg.get("name", ""),
-                        "response": content,
-                    }
+                    response_payload = (
+                        content if isinstance(content, dict) else {"result": content}
+                    )
                     gemini_messages.append(
                         types.Content(
                             role="function",
                             parts=[
-                                types.Part.from_function_response(function_response)  # type: ignore[misc,call-arg,arg-type]
+                                types.Part.from_function_response(
+                                    name=msg.get("name")
+                                    or msg.get("tool_call_id")
+                                    or "",
+                                    response=response_payload,
+                                )
                             ],
                         )
                     )
@@ -343,22 +379,44 @@ class GeminiClient(GenAIClient):
                             )
                         )
                 elif role == "assistant":
-                    gemini_messages.append(
-                        types.Content(
-                            role="model", parts=[types.Part.from_text(text=content)]
-                        )
-                    )
+                    parts: list[types.Part] = []
+                    if content:
+                        parts.append(types.Part.from_text(text=content))
+                    for tc in msg.get("tool_calls") or []:
+                        func = tc.get("function") or {}
+                        tc_name = func.get("name") or ""
+                        tc_args: Any = func.get("arguments")
+                        if isinstance(tc_args, str):
+                            try:
+                                tc_args = json.loads(tc_args)
+                            except (json.JSONDecodeError, TypeError):
+                                tc_args = {}
+                        if not isinstance(tc_args, dict):
+                            tc_args = {}
+                        if tc_name:
+                            parts.append(
+                                types.Part.from_function_call(
+                                    name=tc_name, args=tc_args
+                                )
+                            )
+                    if not parts:
+                        parts.append(types.Part.from_text(text=" "))
+                    gemini_messages.append(types.Content(role="model", parts=parts))
                 elif role == "tool":
                     # Handle tool response
-                    function_response = {
-                        "name": msg.get("name", ""),
-                        "response": content,
-                    }
+                    response_payload = (
+                        content if isinstance(content, dict) else {"result": content}
+                    )
                     gemini_messages.append(
                         types.Content(
                             role="function",
                             parts=[
-                                types.Part.from_function_response(function_response)  # type: ignore[misc,call-arg,arg-type]
+                                types.Part.from_function_response(
+                                    name=msg.get("name")
+                                    or msg.get("tool_call_id")
+                                    or "",
+                                    response=response_payload,
+                                )
                             ],
                         )
                     )
@@ -427,6 +485,7 @@ class GeminiClient(GenAIClient):
             content_parts: list[str] = []
             tool_calls_by_index: dict[int, dict[str, Any]] = {}
             finish_reason = "stop"
+            usage_stats: Optional[dict[str, Any]] = None
 
             stream = await self.provider.aio.models.generate_content_stream(
                 model=self.genai_config.model,
@@ -435,6 +494,12 @@ class GeminiClient(GenAIClient):
             )
 
             async for chunk in stream:
+                chunk_usage = getattr(chunk, "usage_metadata", None)
+                if chunk_usage is not None:
+                    maybe_stats = _stats_from_gemini_usage(chunk_usage)
+                    if maybe_stats is not None:
+                        usage_stats = maybe_stats
+
                 if not chunk or not chunk.candidates:
                     continue
 
@@ -521,6 +586,9 @@ class GeminiClient(GenAIClient):
                     )
                 finish_reason = "tool_calls"
 
+            if usage_stats is not None:
+                yield ("stats", usage_stats)
+
             yield (
                 "message",
                 {

frigate/genai/llama_cpp.py

@@ -18,6 +18,63 @@ from frigate.genai.utils import parse_tool_calls_from_message
 logger = logging.getLogger(__name__)
 
 
+def _stats_from_llama_cpp_chunk(data: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Build a stats dict from a llama.cpp streaming chunk.
+
+    Final-chunk `usage` carries authoritative token counts. Per-chunk
+    `timings` (enabled via timings_per_token) carries the running token
+    counts (prompt_n, predicted_n) and generation rate, so live updates
+    work mid-stream.
+    """
+    usage = data.get("usage") or {}
+    timings = data.get("timings") or {}
+    prompt_tokens = usage.get("prompt_tokens")
+    completion_tokens = usage.get("completion_tokens")
+    predicted_ms = timings.get("predicted_ms")
+    tps = timings.get("predicted_per_second")
+    stats: dict[str, Any] = {}
+
+    if not isinstance(prompt_tokens, int):
+        prompt_n = timings.get("prompt_n")
+
+        if isinstance(prompt_n, int):
+            prompt_tokens = prompt_n
+
+    if not isinstance(completion_tokens, int):
+        predicted_n = timings.get("predicted_n")
+
+        if isinstance(predicted_n, int):
+            completion_tokens = predicted_n
+
+    if not isinstance(prompt_tokens, int) and not isinstance(completion_tokens, int):
+        return None
+
+    if isinstance(prompt_tokens, int):
+        stats["prompt_tokens"] = prompt_tokens
+
+    if isinstance(completion_tokens, int):
+        stats["completion_tokens"] = completion_tokens
+
+    if isinstance(predicted_ms, (int, float)) and predicted_ms > 0:
+        stats["completion_duration_ms"] = float(predicted_ms)
+
+    if isinstance(tps, (int, float)) and tps > 0:
+        stats["tokens_per_second"] = float(tps)
+
+    return stats or None
+
+
+def _parse_launch_arg(args: list[str], flag: str) -> str | None:
+    """Return the value following `flag` in a positional argv list, or None."""
+    try:
+        idx = args.index(flag)
+    except ValueError:
+        return None
+    if idx + 1 >= len(args):
+        return None
+    return args[idx + 1]
+
+
 def _to_jpeg(img_bytes: bytes) -> bytes | None:
     """Convert image bytes to JPEG. llama.cpp/STB does not support WebP."""
     try:
@@ -42,6 +99,9 @@ class LlamaCppClient(GenAIClient):
     _supports_vision: bool
     _supports_audio: bool
     _supports_tools: bool
+    _image_token_cache: dict[tuple[int, int], int]
+    _text_baseline_tokens: int | None
+    _media_marker: str
 
     def _init_provider(self) -> str | None:
         """Initialize the client and query model metadata from the server."""
@@ -52,6 +112,9 @@ class LlamaCppClient(GenAIClient):
         self._supports_vision = False
         self._supports_audio = False
         self._supports_tools = False
+        self._image_token_cache = {}
+        self._text_baseline_tokens = None
+        self._media_marker = "<__media__>"
 
         base_url = (
             self.genai_config.base_url.rstrip("/")
@@ -61,28 +124,73 @@ class LlamaCppClient(GenAIClient):
 
         if base_url is None:
             return None
+        else:
+            base_url = base_url.replace("/v1", "")  # Strip /v1 if included in base_url
 
         configured_model = self.genai_config.model
+        info = self._get_model_info(base_url, configured_model)
 
-        # Query /v1/models to validate the configured model exists
+        if info is None:
+            return None
+
+        self._context_size = info["context_size"]
+        self._supports_vision = info["supports_vision"]
+        self._supports_audio = info["supports_audio"]
+        self._supports_tools = info["supports_tools"]
+        self._media_marker = info["media_marker"]
+
+        logger.info(
+            "llama.cpp model '%s' initialized — context: %s, vision: %s, audio: %s, tools: %s",
+            configured_model,
+            self._context_size or "unknown",
+            self._supports_vision,
+            self._supports_audio,
+            self._supports_tools,
+        )
+
+        return base_url
+
+    def _get_model_info(
+        self, base_url: str, configured_model: str
+    ) -> dict[str, Any] | None:
+        """Resolve model metadata from /v1/models with /props fallback.
+
+        Returns a dict of capability fields, or None if the server's model
+        registry was reachable and reported the configured model as missing.
+        A reachable-but-unparseable /v1/models is treated as soft-pass and
+        falls through to /props, matching prior behavior.
+
+        After ggml-org/llama.cpp#22952, /v1/models exposes per-model
+        `architecture.input_modalities` (text/image/audio) — the primary
+        source. When proxied through llama-swap, the same entry carries
+        `status.args` (server launch argv) and, for the loaded model,
+        `meta.n_ctx`. /props remains the only source for `media_marker`,
+        which the server randomizes per startup unless LLAMA_MEDIA_MARKER
+        is set.
+        """
+        info: dict[str, Any] = {
+            "context_size": None,
+            "supports_vision": False,
+            "supports_audio": False,
+            "supports_tools": False,
+            "media_marker": "<__media__>",
+        }
+
+        model_entry: dict[str, Any] | None = None
         try:
-            response = requests.get(
-                f"{base_url}/v1/models",
-                timeout=10,
-            )
+            response = requests.get(f"{base_url}/v1/models", timeout=10)
             response.raise_for_status()
             models_data = response.json()
 
-            model_found = False
             for model in models_data.get("data", []):
                 model_ids = {model.get("id")}
                 for alias in model.get("aliases", []):
                     model_ids.add(alias)
                 if configured_model in model_ids:
-                    model_found = True
+                    model_entry = model
                     break
 
-            if not model_found:
+            if model_entry is None:
                 available = []
                 for m in models_data.get("data", []):
                     available.append(m.get("id", "unknown"))
@@ -101,10 +209,35 @@ class LlamaCppClient(GenAIClient):
                 e,
             )
 
-        # Query /props for context size, modalities, and tool support.
-        # The standard /props?model=<name> endpoint works with llama-server.
-        # If it fails, try the llama-swap per-model passthrough endpoint which
-        # returns props for a specific model without requiring it to be loaded.
+        if model_entry is not None:
+            architecture = model_entry.get("architecture") or {}
+            input_modalities = architecture.get("input_modalities") or []
+
+            if isinstance(input_modalities, list):
+                info["supports_vision"] = "image" in input_modalities
+                info["supports_audio"] = "audio" in input_modalities
+
+            status = model_entry.get("status") or {}
+            launch_args = status.get("args") if isinstance(status, dict) else None
+            if not isinstance(launch_args, list):
+                launch_args = []
+
+            meta = model_entry.get("meta") if isinstance(model_entry, dict) else None
+            n_ctx = meta.get("n_ctx") if isinstance(meta, dict) else None
+
+            if not n_ctx:
+                n_ctx = _parse_launch_arg(launch_args, "--ctx-size")
+
+            if n_ctx:
+                try:
+                    info["context_size"] = int(n_ctx)
+                except (TypeError, ValueError):
+                    pass
+
+            # Tool calling on llama-server requires --jinja.
+            if "--jinja" in launch_args:
+                info["supports_tools"] = True
+
         try:
             try:
                 response = requests.get(
@@ -122,37 +255,32 @@ class LlamaCppClient(GenAIClient):
                 response.raise_for_status()
                 props = response.json()
 
-            # Context size from server runtime config
-            default_settings = props.get("default_generation_settings", {})
-            n_ctx = default_settings.get("n_ctx")
-            if n_ctx:
-                self._context_size = int(n_ctx)
+            if info["context_size"] is None:
+                default_settings = props.get("default_generation_settings", {})
+                n_ctx = default_settings.get("n_ctx")
+                if n_ctx:
+                    info["context_size"] = int(n_ctx)
 
-            # Modalities (vision, audio)
-            modalities = props.get("modalities", {})
-            self._supports_vision = modalities.get("vision", False)
-            self._supports_audio = modalities.get("audio", False)
+            if not (info["supports_vision"] or info["supports_audio"]):
+                modalities = props.get("modalities", {})
+                info["supports_vision"] = bool(modalities.get("vision", False))
+                info["supports_audio"] = bool(modalities.get("audio", False))
 
-            # Tool support from chat template capabilities
-            chat_caps = props.get("chat_template_caps", {})
-            self._supports_tools = chat_caps.get("supports_tools", False)
+            if not info["supports_tools"]:
+                chat_caps = props.get("chat_template_caps", {})
+                info["supports_tools"] = bool(chat_caps.get("supports_tools", False))
 
-            logger.info(
-                "llama.cpp model '%s' initialized — context: %s, vision: %s, audio: %s, tools: %s",
-                configured_model,
-                self._context_size or "unknown",
-                self._supports_vision,
-                self._supports_audio,
-                self._supports_tools,
-            )
+            media_marker = props.get("media_marker")
+            if isinstance(media_marker, str) and media_marker:
+                info["media_marker"] = media_marker
         except Exception as e:
             logger.warning(
                 "Failed to query llama.cpp /props endpoint: %s. "
-                "Using defaults for context size and capabilities.",
+                "Image embeddings may fail if the server randomized its media marker.",
                 e,
             )
 
-        return base_url
+        return info
 
     def _send(
         self,
@@ -272,6 +400,91 @@ class LlamaCppClient(GenAIClient):
             return self._context_size
         return 4096
 
+    def estimate_image_tokens(self, width: int, height: int) -> float:
+        """Probe the llama.cpp server to learn the model's image-token cost at the
+        requested dimensions.
+
+        llama.cpp's image tokenization is a deterministic function of dimensions and
+        the loaded mmproj, so the result is cached per (width, height) for the
+        lifetime of the process. Falls back to the base pixel heuristic if the
+        server is unreachable or the response is malformed.
+        """
+        if self.provider is None:
+            return super().estimate_image_tokens(width, height)
+
+        cached = self._image_token_cache.get((width, height))
+
+        if cached is not None:
+            return cached
+
+        try:
+            baseline = self._probe_baseline_tokens()
+            with_image = self._probe_image_prompt_tokens(width, height)
+            tokens = max(1, with_image - baseline)
+        except Exception as e:
+            logger.debug(
+                "llama.cpp image-token probe failed for %dx%d (%s); using heuristic",
+                width,
+                height,
+                e,
+            )
+            return super().estimate_image_tokens(width, height)
+
+        self._image_token_cache[(width, height)] = tokens
+        logger.debug(
+            "llama.cpp model '%s' uses ~%d tokens for %dx%d images",
+            self.genai_config.model,
+            tokens,
+            width,
+            height,
+        )
+        return tokens
+
+    def _probe_baseline_tokens(self) -> int:
+        """Return prompt_tokens for a minimal text-only request. Cached after first call."""
+        if self._text_baseline_tokens is not None:
+            return self._text_baseline_tokens
+
+        self._text_baseline_tokens = self._probe_prompt_tokens(
+            [{"type": "text", "text": "."}]
+        )
+        return self._text_baseline_tokens
+
+    def _probe_image_prompt_tokens(self, width: int, height: int) -> int:
+        """Return prompt_tokens for a single synthetic image plus minimal text."""
+        img = Image.new("RGB", (width, height), (128, 128, 128))
+        buf = io.BytesIO()
+        img.save(buf, format="JPEG", quality=60)
+        encoded = base64.b64encode(buf.getvalue()).decode("utf-8")
+        return self._probe_prompt_tokens(
+            [
+                {"type": "text", "text": "."},
+                {
+                    "type": "image_url",
+                    "image_url": {"url": f"data:image/jpeg;base64,{encoded}"},
+                },
+            ]
+        )
+
+    def _probe_prompt_tokens(self, content: list[dict[str, Any]]) -> int:
+        """POST a 1-token chat completion and return reported prompt_tokens.
+
+        Uses a generous timeout to absorb a cold model load on the first probe
+        when the server lazily loads models on demand (e.g. llama-swap).
+        """
+        payload = {
+            "model": self.genai_config.model,
+            "messages": [{"role": "user", "content": content}],
+            "max_tokens": 1,
+        }
+        response = requests.post(
+            f"{self.provider}/v1/chat/completions",
+            json=payload,
+            timeout=60,
+        )
+        response.raise_for_status()
+        return int(response.json()["usage"]["prompt_tokens"])
+
     def _build_payload(
         self,
         messages: list[dict[str, Any]],
@@ -295,6 +508,8 @@ class LlamaCppClient(GenAIClient):
         }
         if stream:
             payload["stream"] = True
+            payload["stream_options"] = {"include_usage": True}
+            payload["timings_per_token"] = True
         if tools:
             payload["tools"] = tools
             if openai_tool_choice is not None:
@@ -376,10 +591,11 @@ class LlamaCppClient(GenAIClient):
             jpeg_bytes = _to_jpeg(img)
             to_encode = jpeg_bytes if jpeg_bytes is not None else img
             encoded = base64.b64encode(to_encode).decode("utf-8")
-            # prompt_string must contain <__media__> placeholder for image tokenization
+            # prompt_string must contain the server's media marker placeholder.
+            # The marker is randomized per server startup (read from /props).
             content.append(
                 {
-                    "prompt_string": "<__media__>\n",
+                    "prompt_string": f"{self._media_marker}\n",
                     "multimodal_data": [encoded],  # type: ignore[dict-item]
                 }
             )
@@ -556,6 +772,9 @@ class LlamaCppClient(GenAIClient):
                             data = json.loads(data_str)
                         except json.JSONDecodeError:
                             continue
+                        maybe_stats = _stats_from_llama_cpp_chunk(data)
+                        if maybe_stats is not None:
+                            yield ("stats", maybe_stats)
                         choices = data.get("choices") or []
                         if not choices:
                             continue

frigate/genai/ollama.py

@@ -1,5 +1,7 @@
 """Ollama Provider for Frigate AI."""
 
+import base64
+import binascii
 import json
 import logging
 from typing import Any, AsyncGenerator, Optional
@@ -16,6 +18,72 @@ from frigate.genai.utils import parse_tool_calls_from_message
 logger = logging.getLogger(__name__)
 
 
+def _extract_ollama_stats(response: Any) -> Optional[dict[str, Any]]:
+    """Build a stats dict from Ollama's response metadata.
+
+    Ollama reports eval_count/eval_duration (generation) and
+    prompt_eval_count (context size). Durations are nanoseconds.
+    """
+    if not response:
+        return None
+    if hasattr(response, "get"):
+        getter = response.get
+    else:
+        getter = lambda key: getattr(response, key, None)  # noqa: E731
+
+    eval_count = getter("eval_count")
+    eval_duration_ns = getter("eval_duration")
+    prompt_eval_count = getter("prompt_eval_count")
+    if eval_count is None and prompt_eval_count is None:
+        return None
+
+    stats: dict[str, Any] = {}
+    if isinstance(prompt_eval_count, int):
+        stats["prompt_tokens"] = prompt_eval_count
+    if isinstance(eval_count, int):
+        stats["completion_tokens"] = eval_count
+    if isinstance(eval_duration_ns, int) and eval_duration_ns > 0:
+        stats["completion_duration_ms"] = eval_duration_ns / 1_000_000
+        if isinstance(eval_count, int) and eval_count > 0:
+            stats["tokens_per_second"] = eval_count / (eval_duration_ns / 1_000_000_000)
+    return stats or None
+
+
+def _normalize_multimodal_content(
+    content: Any,
+) -> tuple[Optional[str], Optional[list[bytes]]]:
+    """Convert OpenAI-style multimodal content to Ollama's (text, images) shape.
+
+    The chat API constructs user messages with content as a list of
+    ``{"type": "text"}`` and ``{"type": "image_url"}`` parts when a tool
+    returns a live frame. Ollama's SDK requires content to be a string and
+    images to be passed in a separate field, so we extract each.
+    """
+    if not isinstance(content, list):
+        return content, None
+
+    text_parts: list[str] = []
+    images: list[bytes] = []
+    for part in content:
+        if not isinstance(part, dict):
+            continue
+        part_type = part.get("type")
+        if part_type == "text":
+            text = part.get("text")
+            if text:
+                text_parts.append(str(text))
+        elif part_type == "image_url":
+            url = (part.get("image_url") or {}).get("url", "")
+            if isinstance(url, str) and url.startswith("data:"):
+                try:
+                    encoded = url.split(",", 1)[1]
+                    images.append(base64.b64decode(encoded, validate=True))
+                except (ValueError, IndexError, binascii.Error) as e:
+                    logger.debug("Failed to decode multimodal image url: %s", e)
+
+    return ("\n".join(text_parts) if text_parts else None), (images or None)
+
+
 @register_genai_provider(GenAIProviderEnum.ollama)
 class OllamaClient(GenAIClient):
     """Generative AI client for Frigate using Ollama."""
@@ -31,6 +99,12 @@ class OllamaClient(GenAIClient):
     provider: ApiClient | None
     provider_options: dict[str, Any]
 
+    def _auth_headers(self) -> dict | None:
+        if self.genai_config.api_key:
+            return {"Authorization": "Bearer " + self.genai_config.api_key}
+
+        return None
+
     def _init_provider(self) -> ApiClient | None:
         """Initialize the client."""
         self.provider_options = {
@@ -39,7 +113,11 @@ class OllamaClient(GenAIClient):
         }
 
         try:
-            client = ApiClient(host=self.genai_config.base_url, timeout=self.timeout)
+            client = ApiClient(
+                host=self.genai_config.base_url,
+                timeout=self.timeout,
+                headers=self._auth_headers(),
+            )
             # ensure the model is available locally
             response = client.show(self.genai_config.model)
             if response.get("error"):
@@ -113,6 +191,15 @@ class OllamaClient(GenAIClient):
                 schema = response_format.get("json_schema", {}).get("schema")
                 if schema:
                     ollama_options["format"] = self._clean_schema_for_ollama(schema)
+            logger.debug(
+                "Ollama generate request: model=%s, prompt_len=%s, image_count=%s, "
+                "has_format=%s, options=%s",
+                self.genai_config.model,
+                len(prompt),
+                len(images) if images else 0,
+                "format" in ollama_options,
+                {k: v for k, v in ollama_options.items() if k != "format"},
+            )
             result = self.provider.generate(
                 self.genai_config.model,
                 prompt,
@@ -120,9 +207,24 @@ class OllamaClient(GenAIClient):
                 **ollama_options,
             )
             logger.debug(
-                f"Ollama tokens used: eval_count={result.get('eval_count')}, prompt_eval_count={result.get('prompt_eval_count')}"
+                "Ollama generate response: done=%s, done_reason=%s, eval_count=%s, "
+                "prompt_eval_count=%s, response_len=%s",
+                result.get("done"),
+                result.get("done_reason"),
+                result.get("eval_count"),
+                result.get("prompt_eval_count"),
+                len(result.get("response", "") or ""),
             )
-            return str(result["response"]).strip()
+            response_text = str(result["response"]).strip()
+            if not response_text:
+                logger.warning(
+                    "Ollama returned a blank response for model %s (done_reason=%s, "
+                    "eval_count=%s). Check model output, ensure thinking is disabled.",
+                    self.genai_config.model,
+                    result.get("done_reason"),
+                    result.get("eval_count"),
+                )
+            return response_text
         except (
             TimeoutException,
             ResponseError,
@@ -142,7 +244,9 @@ class OllamaClient(GenAIClient):
                 return []
             try:
                 client = ApiClient(
-                    host=self.genai_config.base_url, timeout=self.timeout
+                    host=self.genai_config.base_url,
+                    timeout=self.timeout,
+                    headers=self._auth_headers(),
                 )
             except Exception:
                 return []
@@ -171,10 +275,13 @@ class OllamaClient(GenAIClient):
         """Build request_messages and params for chat (sync or stream)."""
         request_messages = []
         for msg in messages:
-            msg_dict = {
+            content, images = _normalize_multimodal_content(msg.get("content", ""))
+            msg_dict: dict[str, Any] = {
                 "role": msg.get("role"),
-                "content": msg.get("content", ""),
+                "content": content if content is not None else "",
             }
+            if images:
+                msg_dict["images"] = images
             if msg.get("tool_call_id"):
                 msg_dict["tool_call_id"] = msg["tool_call_id"]
             if msg.get("name"):
@@ -320,12 +427,16 @@ class OllamaClient(GenAIClient):
                 async_client = OllamaAsyncClient(
                     host=self.genai_config.base_url,
                     timeout=self.timeout,
+                    headers=self._auth_headers(),
                 )
                 response = await async_client.chat(**request_params)
                 result = self._message_from_response(response)
                 content = result.get("content")
                 if content:
                     yield ("content_delta", content)
+                stats = _extract_ollama_stats(response)
+                if stats is not None:
+                    yield ("stats", stats)
                 yield ("message", result)
                 return
 
@@ -335,9 +446,11 @@ class OllamaClient(GenAIClient):
             async_client = OllamaAsyncClient(
                 host=self.genai_config.base_url,
                 timeout=self.timeout,
+                headers=self._auth_headers(),
             )
             content_parts: list[str] = []
             final_message: dict[str, Any] | None = None
+            final_chunk: Any = None
             stream = await async_client.chat(**request_params)
             async for chunk in stream:
                 if not chunk or "message" not in chunk:
@@ -348,6 +461,7 @@ class OllamaClient(GenAIClient):
                     content_parts.append(delta)
                     yield ("content_delta", delta)
                 if chunk.get("done"):
+                    final_chunk = chunk
                     full_content = "".join(content_parts).strip() or None
                     final_message = {
                         "content": full_content,
@@ -356,6 +470,10 @@ class OllamaClient(GenAIClient):
                     }
                     break
 
+            stats = _extract_ollama_stats(final_chunk)
+            if stats is not None:
+                yield ("stats", stats)
+
             if final_message is not None:
                 yield ("message", final_message)
             else:

frigate/genai/openai.py

@@ -14,6 +14,22 @@ from frigate.genai import GenAIClient, register_genai_provider
 logger = logging.getLogger(__name__)
 
 
+def _stats_from_openai_usage(usage: Any) -> Optional[dict[str, Any]]:
+    """Build a stats dict from an OpenAI-compatible usage object."""
+    if usage is None:
+        return None
+    prompt_tokens = getattr(usage, "prompt_tokens", None)
+    completion_tokens = getattr(usage, "completion_tokens", None)
+    if prompt_tokens is None and completion_tokens is None:
+        return None
+    stats: dict[str, Any] = {}
+    if isinstance(prompt_tokens, int):
+        stats["prompt_tokens"] = prompt_tokens
+    if isinstance(completion_tokens, int):
+        stats["completion_tokens"] = completion_tokens
+    return stats or None
+
+
 @register_genai_provider(GenAIProviderEnum.openai)
 class OpenAIClient(GenAIClient):
     """Generative AI client for Frigate using OpenAI."""
@@ -73,14 +89,39 @@ class OpenAIClient(GenAIClient):
                 **self.genai_config.runtime_options,
             }
             if response_format:
+                # OpenAI strict mode requires additionalProperties: false on the schema
+                if response_format.get("type") == "json_schema" and response_format.get(
+                    "json_schema", {}
+                ).get("strict"):
+                    schema = response_format.get("json_schema", {}).get("schema")
+                    if isinstance(schema, dict):
+                        schema["additionalProperties"] = False
                 request_params["response_format"] = response_format
+
             result = self.provider.chat.completions.create(**request_params)
+
             if (
                 result is not None
                 and hasattr(result, "choices")
                 and len(result.choices) > 0
             ):
-                return str(result.choices[0].message.content.strip())
+                message = result.choices[0].message
+                content = message.content
+
+                if not content:
+                    # When reasoning is enabled for some OpenAI backends the actual response
+                    # is incorrectly placed in reasoning_content instead of content.
+                    # This is buggy/incorrect behavior — reasoning should not be
+                    # enabled for these models.
+                    reasoning_content = getattr(message, "reasoning_content", None)
+                    if reasoning_content:
+                        logger.warning(
+                            "Response content was empty but reasoning_content was provided; "
+                            "reasoning appears to be enabled and should be disabled for this model."
+                        )
+                        content = reasoning_content
+
+                return str(content.strip()) if content else None
             return None
         except (TimeoutException, Exception) as e:
             logger.warning("OpenAI returned an error: %s", str(e))
@@ -273,6 +314,7 @@ class OpenAIClient(GenAIClient):
                 "messages": messages,
                 "timeout": self.timeout,
                 "stream": True,
+                "stream_options": {"include_usage": True},
             }
 
             if tools:
@@ -293,10 +335,15 @@ class OpenAIClient(GenAIClient):
             content_parts: list[str] = []
             tool_calls_by_index: dict[int, dict[str, Any]] = {}
             finish_reason = "stop"
+            usage_stats: Optional[dict[str, Any]] = None
 
             stream = self.provider.chat.completions.create(**request_params)  # type: ignore[call-overload]
 
             for chunk in stream:
+                chunk_usage = getattr(chunk, "usage", None)
+                if chunk_usage is not None:
+                    usage_stats = _stats_from_openai_usage(chunk_usage)
+
                 if not chunk or not chunk.choices:
                     continue
 
@@ -356,6 +403,9 @@ class OpenAIClient(GenAIClient):
                     )
                 finish_reason = "tool_calls"
 
+            if usage_stats is not None:
+                yield ("stats", usage_stats)
+
             yield (
                 "message",
                 {

frigate/jobs/debug_replay.py

@@ -0,0 +1,386 @@
+"""Debug replay startup job: ffmpeg concat + camera config publish.
+
+The runner orchestrates the async portion of starting a debug replay
+session. The DebugReplayManager (in frigate.debug_replay) owns session
+presence so the status bar can keep reading a single `active` flag from
+/debug_replay/status for the entire session window — which is broader
+than this job's lifetime.
+"""
+
+import logging
+import os
+import subprocess as sp
+import threading
+import time
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, Optional, cast
+
+from peewee import ModelSelect
+
+from frigate.config import FrigateConfig
+from frigate.config.camera.updater import CameraConfigUpdatePublisher
+from frigate.const import REPLAY_CAMERA_PREFIX, REPLAY_DIR
+from frigate.jobs.export import JobStatePublisher
+from frigate.jobs.job import Job
+from frigate.jobs.manager import job_is_running, set_current_job
+from frigate.models import Recordings
+from frigate.types import JobStatusTypesEnum
+from frigate.util.ffmpeg import run_ffmpeg_with_progress
+
+if TYPE_CHECKING:
+    from frigate.debug_replay import DebugReplayManager
+
+logger = logging.getLogger(__name__)
+
+# Coalesce frequent ffmpeg progress callbacks so the WS isn't flooded.
+PROGRESS_BROADCAST_MIN_INTERVAL = 1.0
+
+JOB_TYPE = "debug_replay"
+
+STEP_PREPARING_CLIP = "preparing_clip"
+STEP_STARTING_CAMERA = "starting_camera"
+
+
+_active_runner: Optional["DebugReplayJobRunner"] = None
+_runner_lock = threading.Lock()
+
+
+def _set_active_runner(runner: Optional["DebugReplayJobRunner"]) -> None:
+    global _active_runner
+    with _runner_lock:
+        _active_runner = runner
+
+
+def get_active_runner() -> Optional["DebugReplayJobRunner"]:
+    with _runner_lock:
+        return _active_runner
+
+
+@dataclass
+class DebugReplayJob(Job):
+    """Job state for a debug replay startup."""
+
+    job_type: str = JOB_TYPE
+    source_camera: str = ""
+    replay_camera_name: str = ""
+    start_ts: float = 0.0
+    end_ts: float = 0.0
+    current_step: Optional[str] = None
+    progress_percent: float = 0.0
+
+    def to_dict(self) -> dict[str, Any]:
+        """Whitelisted payload for the job_state WS topic.
+
+        Replay-specific fields land in results so the frontend's
+        generic Job<TResults> type can be parameterised cleanly.
+        """
+        return {
+            "id": self.id,
+            "job_type": self.job_type,
+            "status": self.status,
+            "start_time": self.start_time,
+            "end_time": self.end_time,
+            "error_message": self.error_message,
+            "results": {
+                "current_step": self.current_step,
+                "progress_percent": self.progress_percent,
+                "source_camera": self.source_camera,
+                "replay_camera_name": self.replay_camera_name,
+                "start_ts": self.start_ts,
+                "end_ts": self.end_ts,
+            },
+        }
+
+
+def query_recordings(source_camera: str, start_ts: float, end_ts: float) -> ModelSelect:
+    """Return the Recordings query for the time range.
+
+    Module-level so tests can patch it without instantiating a runner.
+    """
+    query = (
+        Recordings.select(
+            Recordings.path,
+            Recordings.start_time,
+            Recordings.end_time,
+        )
+        .where(
+            Recordings.start_time.between(start_ts, end_ts)
+            | Recordings.end_time.between(start_ts, end_ts)
+            | ((start_ts > Recordings.start_time) & (end_ts < Recordings.end_time))
+        )
+        .where(Recordings.camera == source_camera)
+        .order_by(Recordings.start_time.asc())
+    )
+    return cast(ModelSelect, query)
+
+
+class DebugReplayJobRunner(threading.Thread):
+    """Worker thread that drives the startup job to completion.
+
+    Owns the live ffmpeg Popen reference for cancellation. Cancellation
+    is two-step (threading.Event + proc.terminate()) so the runner
+    both knows it should stop and is unblocked from its blocking subprocess
+    wait.
+    """
+
+    def __init__(
+        self,
+        job: DebugReplayJob,
+        frigate_config: FrigateConfig,
+        config_publisher: CameraConfigUpdatePublisher,
+        replay_manager: "DebugReplayManager",
+        publisher: Optional[JobStatePublisher] = None,
+    ) -> None:
+        super().__init__(daemon=True, name=f"debug_replay_{job.id}")
+        self.job = job
+        self.frigate_config = frigate_config
+        self.config_publisher = config_publisher
+        self.replay_manager = replay_manager
+        self.publisher = publisher if publisher is not None else JobStatePublisher()
+        self._cancel_event = threading.Event()
+        self._active_process: sp.Popen | None = None
+        self._proc_lock = threading.Lock()
+        self._last_broadcast_monotonic: float = 0.0
+
+    def cancel(self) -> None:
+        """Request cancellation. Idempotent."""
+        self._cancel_event.set()
+        with self._proc_lock:
+            proc = self._active_process
+        if proc is not None:
+            try:
+                proc.terminate()
+            except Exception as exc:
+                logger.warning("Failed to terminate ffmpeg subprocess: %s", exc)
+
+    def is_cancelled(self) -> bool:
+        return self._cancel_event.is_set()
+
+    def _record_proc(self, proc: sp.Popen) -> None:
+        with self._proc_lock:
+            self._active_process = proc
+        # Race: cancel arrived between Popen and _record_proc.
+        if self._cancel_event.is_set():
+            try:
+                proc.terminate()
+            except Exception:
+                pass
+
+    def _broadcast(self, force: bool = False) -> None:
+        now = time.monotonic()
+        if (
+            not force
+            and now - self._last_broadcast_monotonic < PROGRESS_BROADCAST_MIN_INTERVAL
+        ):
+            return
+        self._last_broadcast_monotonic = now
+
+        try:
+            self.publisher.publish(self.job.to_dict())
+        except Exception as err:
+            logger.warning("Publisher raised during job state broadcast: %s", err)
+
+    def run(self) -> None:
+        replay_name = self.job.replay_camera_name
+        os.makedirs(REPLAY_DIR, exist_ok=True)
+        concat_file = os.path.join(REPLAY_DIR, f"{replay_name}_concat.txt")
+        clip_path = os.path.join(REPLAY_DIR, f"{replay_name}.mp4")
+
+        self.job.status = JobStatusTypesEnum.running
+        self.job.start_time = time.time()
+        self.job.current_step = STEP_PREPARING_CLIP
+        self._broadcast(force=True)
+
+        try:
+            recordings = query_recordings(
+                self.job.source_camera, self.job.start_ts, self.job.end_ts
+            )
+            with open(concat_file, "w") as f:
+                for recording in recordings:
+                    f.write(f"file '{recording.path}'\n")
+
+            ffmpeg_cmd = [
+                self.frigate_config.ffmpeg.ffmpeg_path,
+                "-hide_banner",
+                "-y",
+                "-f",
+                "concat",
+                "-safe",
+                "0",
+                "-i",
+                concat_file,
+                "-c",
+                "copy",
+                "-movflags",
+                "+faststart",
+                clip_path,
+            ]
+
+            logger.info(
+                "Generating replay clip for %s (%.1f - %.1f)",
+                self.job.source_camera,
+                self.job.start_ts,
+                self.job.end_ts,
+            )
+
+            def _on_progress(percent: float) -> None:
+                self.job.progress_percent = percent
+                self._broadcast()
+
+            try:
+                returncode, stderr = run_ffmpeg_with_progress(
+                    ffmpeg_cmd,
+                    expected_duration_seconds=max(
+                        0.0, self.job.end_ts - self.job.start_ts
+                    ),
+                    on_progress=_on_progress,
+                    process_started=self._record_proc,
+                    use_low_priority=True,
+                )
+            finally:
+                with self._proc_lock:
+                    self._active_process = None
+
+            if self._cancel_event.is_set():
+                self._finalize_cancelled(clip_path)
+                return
+
+            if returncode != 0:
+                raise RuntimeError(f"FFmpeg failed: {stderr[-500:]}")
+
+            if not os.path.exists(clip_path):
+                raise RuntimeError("Clip file was not created")
+
+            self.job.current_step = STEP_STARTING_CAMERA
+            self.job.progress_percent = 100.0
+            self._broadcast(force=True)
+
+            if self._cancel_event.is_set():
+                self._finalize_cancelled(clip_path)
+                return
+
+            self.replay_manager.publish_camera(
+                source_camera=self.job.source_camera,
+                replay_name=replay_name,
+                clip_path=clip_path,
+                frigate_config=self.frigate_config,
+                config_publisher=self.config_publisher,
+            )
+            self.replay_manager.mark_session_ready(clip_path)
+
+            self.job.status = JobStatusTypesEnum.success
+            self.job.end_time = time.time()
+            self._broadcast(force=True)
+            logger.info(
+                "Debug replay started: %s -> %s",
+                self.job.source_camera,
+                replay_name,
+            )
+        except Exception as exc:
+            logger.exception("Debug replay startup failed")
+            self.job.status = JobStatusTypesEnum.failed
+            self.job.error_message = str(exc)
+            self.job.end_time = time.time()
+            self._broadcast(force=True)
+            self.replay_manager.clear_session()
+            _remove_silent(clip_path)
+        finally:
+            _remove_silent(concat_file)
+            _set_active_runner(None)
+
+    def _finalize_cancelled(self, clip_path: str) -> None:
+        logger.info("Debug replay startup cancelled")
+        self.job.status = JobStatusTypesEnum.cancelled
+        self.job.end_time = time.time()
+        self._broadcast(force=True)
+        # The caller of cancel_debug_replay_job (DebugReplayManager.stop) owns
+        # session cleanup — db rows, filesystem artifacts, clear_session. We
+        # only clean up the partial concat output we created.
+        _remove_silent(clip_path)
+
+
+def _remove_silent(path: str) -> None:
+    try:
+        if os.path.exists(path):
+            os.remove(path)
+    except OSError:
+        pass
+
+
+def start_debug_replay_job(
+    *,
+    source_camera: str,
+    start_ts: float,
+    end_ts: float,
+    frigate_config: FrigateConfig,
+    config_publisher: CameraConfigUpdatePublisher,
+    replay_manager: "DebugReplayManager",
+) -> str:
+    """Validate, create job, start runner. Returns the job id.
+
+    Raises ValueError for bad params (camera missing, time range
+    invalid, no recordings) and RuntimeError if a session is already
+    active.
+    """
+    if job_is_running(JOB_TYPE) or replay_manager.active:
+        raise RuntimeError("A replay session is already active")
+
+    if source_camera not in frigate_config.cameras:
+        raise ValueError(f"Camera '{source_camera}' not found")
+
+    if end_ts <= start_ts:
+        raise ValueError("End time must be after start time")
+
+    recordings = query_recordings(source_camera, start_ts, end_ts)
+    if not recordings.count():
+        raise ValueError(
+            f"No recordings found for camera '{source_camera}' in the specified time range"
+        )
+
+    replay_name = f"{REPLAY_CAMERA_PREFIX}{source_camera}"
+    replay_manager.mark_starting(
+        source_camera=source_camera,
+        replay_camera_name=replay_name,
+        start_ts=start_ts,
+        end_ts=end_ts,
+    )
+
+    job = DebugReplayJob(
+        source_camera=source_camera,
+        replay_camera_name=replay_name,
+        start_ts=start_ts,
+        end_ts=end_ts,
+    )
+    set_current_job(job)
+
+    runner = DebugReplayJobRunner(
+        job=job,
+        frigate_config=frigate_config,
+        config_publisher=config_publisher,
+        replay_manager=replay_manager,
+    )
+    _set_active_runner(runner)
+    runner.start()
+
+    return job.id
+
+
+def cancel_debug_replay_job() -> bool:
+    """Signal the active runner to cancel.
+
+    Returns True if a runner was signalled, False if no job was active.
+    """
+    runner = get_active_runner()
+    if runner is None:
+        return False
+    runner.cancel()
+    return True
+
+
+def wait_for_runner(timeout: float = 2.0) -> bool:
+    """Join the active runner. Returns True if the runner ended in time."""
+    runner = get_active_runner()
+    if runner is None:
+        return True
+    runner.join(timeout=timeout)
+    return not runner.is_alive()

frigate/jobs/vlm_watch.py

@@ -45,6 +45,7 @@ class VLMWatchJob(Job):
     last_reasoning: str = ""
     notification_message: str = ""
     iteration_count: int = 0
+    username: str = ""
 
     def to_dict(self) -> dict[str, Any]:
         return asdict(self)
@@ -374,6 +375,7 @@ def start_vlm_watch_job(
     dispatcher: Any,
     labels: list[str] | None = None,
     zones: list[str] | None = None,
+    username: str = "",
 ) -> str:
     """Start a new VLM watch job. Returns the job ID.
 
@@ -397,6 +399,7 @@ def start_vlm_watch_job(
             max_duration_minutes=max_duration_minutes,
             labels=labels or [],
             zones=zones or [],
+            username=username,
         )
         cancel_ev = threading.Event()
         _current_job = job

frigate/output/birdseye.py

@@ -8,7 +8,6 @@ import os
 import queue
 import subprocess as sp
 import threading
-import time
 import traceback
 from multiprocessing.synchronize import Event as MpEvent
 from typing import Any, Optional
@@ -19,6 +18,7 @@ import numpy as np
 from frigate.comms.inter_process import InterProcessRequestor
 from frigate.config import BirdseyeModeEnum, FfmpegConfig, FrigateConfig
 from frigate.const import BASE_DIR, BIRDSEYE_PIPE, INSTALL_DIR, UPDATE_BIRDSEYE_LAYOUT
+from frigate.output.ws_auth import ws_has_camera_access
 from frigate.util.image import (
     SharedMemoryFrameManager,
     copy_yuv_to_position,
@@ -62,8 +62,10 @@ def get_canvas_shape(width: int, height: int) -> tuple[int, int]:
     if round(a_w / a_h, 2) != round(width / height, 2):
         canvas_width = int(width // 4 * 4)
         canvas_height = int((canvas_width / a_w * a_h) // 4 * 4)
-        logger.warning(
-            f"The birdseye resolution is a non-standard aspect ratio, forcing birdseye resolution to {canvas_width} x {canvas_height}"
+        logger.error(
+            f"Birdseye resolution {width}x{height} is not a supported aspect ratio "
+            f"and may cause visual distortion; falling back to {canvas_width}x{canvas_height}. "
+            f"Set width and height to a supported aspect ratio (16:9, 20:10, 16:6, 32:9, 12:9, 22:15, 9:16, 9:12, 16:3, or 1:1)"
         )
 
     return (canvas_width, canvas_height)
@@ -236,12 +238,14 @@ class BroadcastThread(threading.Thread):
         converter: FFMpegConverter,
         websocket_server: Any,
         stop_event: MpEvent,
+        config: FrigateConfig,
     ):
         super().__init__()
         self.camera = camera
         self.converter = converter
         self.websocket_server = websocket_server
         self.stop_event = stop_event
+        self.config = config
 
     def run(self) -> None:
         while not self.stop_event.is_set():
@@ -256,6 +260,7 @@ class BroadcastThread(threading.Thread):
                     if (
                         not ws.terminated
                         and ws.environ["PATH_INFO"] == f"/{self.camera}"
+                        and ws_has_camera_access(ws, self.camera, self.config)
                     ):
                         try:
                             ws.send(buf, binary=True)
@@ -793,20 +798,27 @@ class Birdseye:
         websocket_server: Any,
     ) -> None:
         self.config = config
+        canvas_width, canvas_height = get_canvas_shape(
+            config.birdseye.width, config.birdseye.height
+        )
         self.input: queue.Queue[bytes] = queue.Queue(maxsize=10)
         self.converter = FFMpegConverter(
             config.ffmpeg,
             self.input,
             stop_event,
-            config.birdseye.width,
-            config.birdseye.height,
-            config.birdseye.width,
-            config.birdseye.height,
+            canvas_width,
+            canvas_height,
+            canvas_width,
+            canvas_height,
             config.birdseye.quality,
             config.birdseye.restream,
         )
         self.broadcaster = BroadcastThread(
-            "birdseye", self.converter, websocket_server, stop_event
+            "birdseye",
+            self.converter,
+            websocket_server,
+            stop_event,
+            config,
         )
         self.birdseye_manager = BirdsEyeFrameManager(self.config, stop_event)
         self.frame_manager = SharedMemoryFrameManager()
@@ -874,7 +886,7 @@ class Birdseye:
                 coordinates = self.birdseye_manager.get_camera_coordinates()
                 self.requestor.send_data(UPDATE_BIRDSEYE_LAYOUT, coordinates)
         if self._idle_interval:
-            now = time.monotonic()
+            now = datetime.datetime.now().timestamp()
             is_idle = len(self.birdseye_manager.camera_layout) == 0
             if (
                 is_idle

frigate/output/camera.py

@@ -7,7 +7,8 @@ import threading
 from multiprocessing.synchronize import Event as MpEvent
 from typing import Any
 
-from frigate.config import CameraConfig, FfmpegConfig
+from frigate.config import CameraConfig, FfmpegConfig, FrigateConfig
+from frigate.output.ws_auth import ws_has_camera_access
 
 logger = logging.getLogger(__name__)
 
@@ -102,12 +103,14 @@ class BroadcastThread(threading.Thread):
         converter: FFMpegConverter,
         websocket_server: Any,
         stop_event: MpEvent,
+        config: FrigateConfig,
     ):
         super().__init__()
         self.camera = camera
         self.converter = converter
         self.websocket_server = websocket_server
         self.stop_event = stop_event
+        self.config = config
 
     def run(self) -> None:
         while not self.stop_event.is_set():
@@ -122,6 +125,7 @@ class BroadcastThread(threading.Thread):
                     if (
                         not ws.terminated
                         and ws.environ["PATH_INFO"] == f"/{self.camera}"
+                        and ws_has_camera_access(ws, self.camera, self.config)
                     ):
                         try:
                             ws.send(buf, binary=True)
@@ -135,7 +139,11 @@ class BroadcastThread(threading.Thread):
 
 class JsmpegCamera:
     def __init__(
-        self, config: CameraConfig, stop_event: MpEvent, websocket_server: Any
+        self,
+        config: CameraConfig,
+        frigate_config: FrigateConfig,
+        stop_event: MpEvent,
+        websocket_server: Any,
     ) -> None:
         self.config = config
         self.input: queue.Queue[bytes] = queue.Queue(maxsize=config.detect.fps)
@@ -154,7 +162,11 @@ class JsmpegCamera:
             config.live.quality,
         )
         self.broadcaster = BroadcastThread(
-            config.name or "", self.converter, websocket_server, stop_event
+            config.name or "",
+            self.converter,
+            websocket_server,
+            stop_event,
+            frigate_config,
         )
 
         self.converter.start()

frigate/output/output.py

@@ -32,6 +32,7 @@ from frigate.const import (
 from frigate.output.birdseye import Birdseye
 from frigate.output.camera import JsmpegCamera
 from frigate.output.preview import PreviewRecorder
+from frigate.output.ws_auth import ws_has_camera_access
 from frigate.util.image import SharedMemoryFrameManager, get_blank_yuv_frame
 from frigate.util.process import FrigateProcess
 
@@ -102,7 +103,7 @@ class OutputProcess(FrigateProcess):
     ) -> None:
         camera_config = self.config.cameras[camera]
         jsmpeg_cameras[camera] = JsmpegCamera(
-            camera_config, self.stop_event, websocket_server
+            camera_config, self.config, self.stop_event, websocket_server
         )
         preview_recorders[camera] = PreviewRecorder(camera_config)
         preview_write_times[camera] = 0
@@ -262,6 +263,7 @@ class OutputProcess(FrigateProcess):
             # send camera frame to ffmpeg process if websockets are connected
             if any(
                 ws.environ["PATH_INFO"].endswith(camera)
+                and ws_has_camera_access(ws, camera, self.config)
                 for ws in websocket_server.manager
             ):
                 # write to the converter for the camera if clients are listening to the specific camera
@@ -275,6 +277,7 @@ class OutputProcess(FrigateProcess):
                     self.config.birdseye.restream
                     or any(
                         ws.environ["PATH_INFO"].endswith("birdseye")
+                        and ws_has_camera_access(ws, "birdseye", self.config)
                         for ws in websocket_server.manager
                     )
                 )
@@ -346,6 +349,13 @@ def move_preview_frames(loc: str) -> None:
             if not os.path.exists(preview_holdover):
                 return
 
+            if not os.access(preview_holdover, os.R_OK | os.W_OK):
+                logger.error(
+                    "Insufficient permissions on preview restart cache at %s",
+                    preview_holdover,
+                )
+                return
+
             shutil.move(preview_holdover, preview_cache)
     except shutil.Error:
         logger.error("Failed to restore preview cache.")

frigate/output/preview.py

@@ -361,14 +361,17 @@ class PreviewRecorder:
             small_frame,
             cv2.COLOR_YUV2BGR_I420,
         )
-        cv2.imwrite(
-            get_cache_image_name(self.camera_name, frame_time),
+        cache_path = get_cache_image_name(self.camera_name, frame_time)
+
+        if not cv2.imwrite(
+            cache_path,
             small_frame,
             [
                 int(cv2.IMWRITE_WEBP_QUALITY),
                 PREVIEW_QUALITY_WEBP[self.config.record.preview.quality],
             ],
-        )
+        ):
+            logger.error("Failed to write preview frame to %s", cache_path)
 
     def write_data(
         self,

frigate/output/ws_auth.py

@@ -0,0 +1,43 @@
+"""Authorization helpers for JSMPEG websocket clients."""
+
+from typing import Any
+
+from frigate.config import FrigateConfig
+from frigate.models import User
+
+
+def _get_valid_ws_roles(ws: Any, config: FrigateConfig) -> list[str]:
+    role_header = ws.environ.get("HTTP_REMOTE_ROLE", "")
+    roles = [
+        role.strip()
+        for role in role_header.split(config.proxy.separator)
+        if role.strip()
+    ]
+    return [role for role in roles if role in config.auth.roles]
+
+
+def ws_has_camera_access(ws: Any, camera_name: str, config: FrigateConfig) -> bool:
+    """Return True when a websocket client is authorized for the camera path."""
+    roles = _get_valid_ws_roles(ws, config)
+
+    if not roles:
+        return False
+
+    roles_dict = config.auth.roles
+
+    # Birdseye is a composite stream, so only users with unrestricted access
+    # should receive it.
+    if camera_name == "birdseye":
+        return any(role == "admin" or not roles_dict.get(role) for role in roles)
+
+    all_camera_names = set(config.cameras.keys())
+
+    for role in roles:
+        if role == "admin" or not roles_dict.get(role):
+            return True
+
+        allowed_cameras = User.get_allowed_cameras(role, roles_dict, all_camera_names)
+        if camera_name in allowed_cameras:
+            return True
+
+    return False

frigate/record/cleanup.py

@@ -351,9 +351,11 @@ class RecordingCleanup(threading.Thread):
                 )
                 .where(
                     ReviewSegment.camera == camera,
-                    # need to ensure segments for all reviews starting
-                    # before the expire date are included
-                    ReviewSegment.start_time < motion_expire_date,
+                    # candidate recordings can extend up to continuous_expire_date
+                    # (the no-motion no-audio branch of the recordings query),
+                    # so reviews must cover that full range to avoid deleting
+                    # segments that overlap recent alerts/detections.
+                    ReviewSegment.start_time < continuous_expire_date,
                 )
                 .order_by(ReviewSegment.start_time)
                 .namedtuples()

frigate/record/export.py

@@ -13,6 +13,7 @@ from enum import Enum
 from pathlib import Path
 from typing import Callable, Optional
 
+import pytz  # type: ignore[import-untyped]
 from peewee import DoesNotExist
 
 from frigate.config import FfmpegConfig, FrigateConfig
@@ -22,13 +23,13 @@ from frigate.const import (
     EXPORT_DIR,
     MAX_PLAYLIST_SECONDS,
     PREVIEW_FRAME_TYPE,
-    PROCESS_PRIORITY_LOW,
 )
 from frigate.ffmpeg_presets import (
     EncodeTypeEnum,
     parse_preset_hardware_acceleration_encode,
 )
-from frigate.models import Export, Previews, Recordings
+from frigate.models import Export, Previews, Recordings, ReviewSegment
+from frigate.util.ffmpeg import run_ffmpeg_with_progress
 from frigate.util.time import is_current_hour
 
 logger = logging.getLogger(__name__)
@@ -242,111 +243,177 @@ class RecordingExporter(threading.Thread):
 
         return total
 
-    def _inject_progress_flags(self, ffmpeg_cmd: list[str]) -> list[str]:
-        """Insert FFmpeg progress reporting flags before the output path.
-
-        ``-progress pipe:2`` writes structured key=value lines to stderr,
-        ``-nostats`` suppresses the noisy default stats output.
-        """
-        if not ffmpeg_cmd:
-            return ffmpeg_cmd
-        return ffmpeg_cmd[:-1] + ["-progress", "pipe:2", "-nostats", ffmpeg_cmd[-1]]
-
     def _run_ffmpeg_with_progress(
         self,
         ffmpeg_cmd: list[str],
         playlist_lines: str | list[str],
         step: str = "encoding",
     ) -> tuple[int, str]:
-        """Run an FFmpeg export command, parsing progress events from stderr.
+        """Delegate to the shared helper, mapping percent → (step, percent).
 
-        Returns ``(returncode, captured_stderr)``. Stdout is left attached to
-        the parent process so we don't have to drain it (and risk a deadlock
-        if the buffer fills). Progress percent is computed against the
-        expected output duration; values are clamped to [0, 100] inside
-        :py:meth:`_emit_progress`.
+        Returns ``(returncode, captured_stderr)``.
         """
-        cmd = ["nice", "-n", str(PROCESS_PRIORITY_LOW)] + self._inject_progress_flags(
-            ffmpeg_cmd
-        )
-
         if isinstance(playlist_lines, list):
             stdin_payload = "\n".join(playlist_lines)
         else:
             stdin_payload = playlist_lines
 
-        expected_duration = self._expected_output_duration_seconds()
-
-        self._emit_progress(step, 0.0)
-
-        proc = sp.Popen(
-            cmd,
-            stdin=sp.PIPE,
-            stderr=sp.PIPE,
-            text=True,
-            encoding="ascii",
-            errors="replace",
+        return run_ffmpeg_with_progress(
+            ffmpeg_cmd,
+            expected_duration_seconds=self._expected_output_duration_seconds(),
+            on_progress=lambda percent: self._emit_progress(step, percent),
+            stdin_payload=stdin_payload,
+            use_low_priority=True,
         )
 
-        assert proc.stdin is not None
-        assert proc.stderr is not None
-
-        try:
-            proc.stdin.write(stdin_payload)
-        except (BrokenPipeError, OSError):
-            # FFmpeg may have rejected the input early; still wait for it
-            # to terminate so the returncode is meaningful.
-            pass
-        finally:
-            try:
-                proc.stdin.close()
-            except (BrokenPipeError, OSError):
-                pass
-
-        captured: list[str] = []
-
-        try:
-            for raw_line in proc.stderr:
-                captured.append(raw_line)
-                line = raw_line.strip()
-
-                if not line:
-                    continue
-
-                if line.startswith("out_time_us="):
-                    if expected_duration <= 0:
-                        continue
-                    try:
-                        out_time_us = int(line.split("=", 1)[1])
-                    except (ValueError, IndexError):
-                        continue
-                    if out_time_us < 0:
-                        continue
-                    out_seconds = out_time_us / 1_000_000.0
-                    percent = (out_seconds / expected_duration) * 100.0
-                    self._emit_progress(step, percent)
-                elif line == "progress=end":
-                    self._emit_progress(step, 100.0)
-                    break
-        except Exception:
-            logger.exception("Failed reading FFmpeg progress for %s", self.export_id)
-
-        proc.wait()
-
-        # Drain any remaining stderr so callers can log it on failure.
-        try:
-            remaining = proc.stderr.read()
-            if remaining:
-                captured.append(remaining)
-        except Exception:
-            pass
-
-        return proc.returncode, "".join(captured)
-
     def get_datetime_from_timestamp(self, timestamp: int) -> str:
-        # return in iso format
+        # return in iso format using the configured ui.timezone when set,
+        # so the auto-generated export name reflects local time rather
+        # than the container's UTC clock
+        tz_name = self.config.ui.timezone
+        if tz_name:
+            try:
+                tz = pytz.timezone(tz_name)
+            except pytz.UnknownTimeZoneError:
+                tz = None
+            if tz is not None:
+                return datetime.datetime.fromtimestamp(timestamp, tz=tz).strftime(
+                    "%Y-%m-%d %H:%M:%S"
+                )
         return datetime.datetime.fromtimestamp(timestamp).strftime("%Y-%m-%d %H:%M:%S")
 
+    def _chapter_metadata_path(self) -> str:
+        return os.path.join(CACHE_DIR, f"export_chapters_{self.export_id}.txt")
+
+    def _build_chapter_metadata_file(self, recordings: list) -> Optional[str]:
+        """Write an FFmpeg metadata file with chapters for review items in range.
+
+        Chapter offsets are computed in *output time*: the VOD endpoint
+        concatenates recording clips back-to-back, so wall-clock gaps
+        between recordings collapse in the produced video. We walk the
+        same recording rows that feed the playlist and convert each
+        review item's wall-clock boundaries into output-time offsets.
+        Returns ``None`` when there are no recordings, no review items,
+        or any chapter would have zero output duration.
+        """
+        if not recordings:
+            return None
+
+        windows: list[tuple[float, float, float]] = []
+        output_offset = 0.0
+        for rec in recordings:
+            clipped_start = max(float(rec.start_time), float(self.start_time))
+            clipped_end = min(float(rec.end_time), float(self.end_time))
+            if clipped_end <= clipped_start:
+                continue
+            windows.append((clipped_start, clipped_end, output_offset))
+            output_offset += clipped_end - clipped_start
+
+        if not windows:
+            return None
+
+        try:
+            review_rows = list(
+                ReviewSegment.select(
+                    ReviewSegment.start_time,
+                    ReviewSegment.end_time,
+                    ReviewSegment.severity,
+                    ReviewSegment.data,
+                )
+                .where(
+                    ReviewSegment.start_time.between(self.start_time, self.end_time)
+                    | ReviewSegment.end_time.between(self.start_time, self.end_time)
+                    | (
+                        (self.start_time > ReviewSegment.start_time)
+                        & (self.end_time < ReviewSegment.end_time)
+                    )
+                )
+                .where(ReviewSegment.camera == self.camera)
+                .order_by(ReviewSegment.start_time.asc())
+                .iterator()
+            )
+        except Exception:
+            logger.exception(
+                "Failed to query review segments for export %s", self.export_id
+            )
+            return None
+
+        if not review_rows:
+            return None
+
+        total_output = windows[-1][2] + (windows[-1][1] - windows[-1][0])
+        last_recorded_end = windows[-1][1]
+
+        def wall_to_output(t: float) -> float:
+            t = max(float(self.start_time), min(float(self.end_time), t))
+            for w_start, w_end, w_offset in windows:
+                if t < w_start:
+                    return w_offset
+                if t <= w_end:
+                    return w_offset + (t - w_start)
+            return total_output
+
+        chapter_blocks: list[str] = []
+        for review in review_rows:
+            if review.start_time is None:
+                continue
+            # In-progress segments have a NULL end_time until the activity
+            # closes; clamp to the last recorded second so the chapter never
+            # extends past the actual video.
+            review_end = (
+                float(review.end_time)
+                if review.end_time is not None
+                else last_recorded_end
+            )
+            start_out = wall_to_output(float(review.start_time))
+            end_out = wall_to_output(review_end)
+
+            # Drop chapters that fall entirely in a recording gap, or are
+            # too short to be navigable in a player.
+            if end_out - start_out < 1.0:
+                continue
+
+            data = review.data or {}
+            labels: list[str] = []
+            for obj in data.get("objects") or []:
+                label = str(obj).split("-")[0]
+                if label and label not in labels:
+                    labels.append(label)
+
+            metadata = data.get("metadata") or {}
+            title = metadata.get("title")
+
+            if not title:
+                title = str(review.severity).capitalize()
+
+                if labels:
+                    title = f"{title}: {', '.join(labels)}"
+
+            chapter_blocks.append(
+                "[CHAPTER]\n"
+                "TIMEBASE=1/1000\n"
+                f"START={int(start_out * 1000)}\n"
+                f"END={int(end_out * 1000)}\n"
+                f"title={title}"
+            )
+
+        if not chapter_blocks:
+            return None
+
+        meta_path = self._chapter_metadata_path()
+        try:
+            with open(meta_path, "w", encoding="utf-8") as f:
+                f.write(";FFMETADATA1\n")
+                f.write("\n".join(chapter_blocks))
+                f.write("\n")
+        except OSError:
+            logger.exception(
+                "Failed to write chapter metadata file for export %s", self.export_id
+            )
+            return None
+
+        return meta_path
+
     def save_thumbnail(self, id: str) -> str:
         thumb_path = os.path.join(CLIPS_DIR, f"export/{id}.webp")
 
@@ -387,16 +454,14 @@ class RecordingExporter(threading.Thread):
             except DoesNotExist:
                 return ""
 
-            diff = self.start_time - preview.start_time
-            minutes = int(diff / 60)
-            seconds = int(diff % 60)
+            diff = max(0.0, float(self.start_time) - float(preview.start_time))
             ffmpeg_cmd = [
                 "/usr/lib/ffmpeg/7.0/bin/ffmpeg",  # hardcode path for exports thumbnail due to missing libwebp support
                 "-hide_banner",
                 "-loglevel",
                 "warning",
                 "-ss",
-                f"00:{minutes}:{seconds}",
+                f"{diff:.3f}",
                 "-i",
                 preview.path,
                 "-frames",
@@ -422,12 +487,18 @@ class RecordingExporter(threading.Thread):
             start_file = f"{file_start}{self.start_time}.{PREVIEW_FRAME_TYPE}"
             end_file = f"{file_start}{self.end_time}.{PREVIEW_FRAME_TYPE}"
             selected_preview = None
+            # Preview frames are written at most 1-2 fps during activity
+            # and as little as one every 30s during quiet periods, so a
+            # short export window can contain zero frames. Track the most
+            # recent frame before the window as a fallback.
+            fallback_preview = None
 
             for file in sorted(os.listdir(preview_dir)):
                 if not file.startswith(file_start):
                     continue
 
                 if file < start_file:
+                    fallback_preview = os.path.join(preview_dir, file)
                     continue
 
                 if file > end_file:
@@ -436,6 +507,9 @@ class RecordingExporter(threading.Thread):
                 selected_preview = os.path.join(preview_dir, file)
                 break
 
+            if not selected_preview:
+                selected_preview = fallback_preview
+
             if not selected_preview:
                 return ""
 
@@ -451,6 +525,24 @@ class RecordingExporter(threading.Thread):
         if type(internal_port) is str:
             internal_port = int(internal_port.split(":")[-1])
 
+        recordings = list(
+            Recordings.select(
+                Recordings.start_time,
+                Recordings.end_time,
+            )
+            .where(
+                Recordings.start_time.between(self.start_time, self.end_time)
+                | Recordings.end_time.between(self.start_time, self.end_time)
+                | (
+                    (self.start_time > Recordings.start_time)
+                    & (self.end_time < Recordings.end_time)
+                )
+            )
+            .where(Recordings.camera == self.camera)
+            .order_by(Recordings.start_time.asc())
+            .iterator()
+        )
+
         playlist_lines: list[str] = []
         if (self.end_time - self.start_time) <= MAX_PLAYLIST_SECONDS:
             playlist_url = f"http://127.0.0.1:{internal_port}/vod/{self.camera}/start/{self.start_time}/end/{self.end_time}/index.m3u8"
@@ -458,32 +550,13 @@ class RecordingExporter(threading.Thread):
                 f"-y -protocol_whitelist pipe,file,http,tcp -i {playlist_url}"
             )
         else:
-            # get full set of recordings
-            export_recordings = (
-                Recordings.select(
-                    Recordings.start_time,
-                    Recordings.end_time,
-                )
-                .where(
-                    Recordings.start_time.between(self.start_time, self.end_time)
-                    | Recordings.end_time.between(self.start_time, self.end_time)
-                    | (
-                        (self.start_time > Recordings.start_time)
-                        & (self.end_time < Recordings.end_time)
-                    )
-                )
-                .where(Recordings.camera == self.camera)
-                .order_by(Recordings.start_time.asc())
-            )
-
-            # Use pagination to process records in chunks
+            # Chunk the recording rows into pages so each playlist line
+            # references a bounded sub-range rather than the full export.
             page_size = 1000
-            num_pages = (export_recordings.count() + page_size - 1) // page_size
-
-            for page in range(1, num_pages + 1):
-                playlist = export_recordings.paginate(page, page_size)
+            for i in range(0, len(recordings), page_size):
+                chunk = recordings[i : i + page_size]
                 playlist_lines.append(
-                    f"file 'http://127.0.0.1:{internal_port}/vod/{self.camera}/start/{float(playlist[0].start_time)}/end/{float(playlist[-1].end_time)}/index.m3u8'"
+                    f"file 'http://127.0.0.1:{internal_port}/vod/{self.camera}/start/{float(chunk[0].start_time)}/end/{float(chunk[-1].end_time)}/index.m3u8'"
                 )
 
             ffmpeg_input = "-y -protocol_whitelist pipe,file,http,tcp -f concat -safe 0 -i /dev/stdin"
@@ -504,8 +577,12 @@ class RecordingExporter(threading.Thread):
                 )
             ).split(" ")
         else:
+            chapters_path = self._build_chapter_metadata_file(recordings)
+            chapter_args = (
+                f" -i {chapters_path} -map 0 -map_metadata 1" if chapters_path else ""
+            )
             ffmpeg_cmd = (
-                f"{self.config.ffmpeg.ffmpeg_path} -hide_banner {ffmpeg_input} -c copy -movflags +faststart"
+                f"{self.config.ffmpeg.ffmpeg_path} -hide_banner {ffmpeg_input}{chapter_args} -c copy -movflags +faststart"
             ).split(" ")
 
         # add metadata
@@ -691,6 +768,8 @@ class RecordingExporter(threading.Thread):
                 ffmpeg_cmd, playlist_lines, step="encoding_retry"
             )
 
+        Path(self._chapter_metadata_path()).unlink(missing_ok=True)
+
         if returncode != 0:
             logger.error(
                 f"Failed to export {self.playback_source.value} for command {' '.join(ffmpeg_cmd)}"

frigate/record/maintainer.py

@@ -610,8 +610,7 @@ class RecordingMaintainer(threading.Thread):
             camera,
         )
 
-        if not os.path.exists(directory):
-            os.makedirs(directory)
+        os.makedirs(directory, exist_ok=True)
 
         # file will be in utc due to start_time being in utc
         file_name = f"{start_time.strftime('%M.%S.mp4')}"

frigate/stats/intel_gpu_info.py

@@ -0,0 +1,109 @@
+"""Resolve human-readable names for Intel GPUs via OpenVINO."""
+
+import logging
+import re
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+class IntelGpuNameResolver:
+    """Build a pdev -> normalized device name map by enumerating OpenVINO GPUs.
+
+    The lookup is performed once on first access and cached for the process
+    lifetime. OpenVINO exposes DEVICE_PCI_INFO (domain/bus/device/function) and
+    FULL_DEVICE_NAME for each GPU it can see, which is enough to associate the
+    name with the pdev string used by DRM fdinfo.
+    """
+
+    _names: Optional[dict[str, str]] = None
+
+    def get_names(self) -> dict[str, str]:
+        if self._names is not None:
+            return self._names
+
+        names: dict[str, str] = {}
+
+        try:
+            from openvino import Core
+        except ImportError:
+            logger.debug("OpenVINO unavailable; cannot resolve Intel GPU names")
+            self._names = names
+            return names
+
+        try:
+            core = Core()
+            devices = core.available_devices
+        except Exception as exc:
+            logger.debug(f"OpenVINO Core initialization failed: {exc}")
+            self._names = names
+            return names
+
+        cpu_name: Optional[str] = None
+        if "CPU" in devices:
+            try:
+                cpu_name = self._strip_trademarks(
+                    core.get_property("CPU", "FULL_DEVICE_NAME")
+                )
+            except Exception as exc:
+                logger.debug(f"Failed to read CPU FULL_DEVICE_NAME: {exc}")
+
+        for device in devices:
+            if not device.startswith("GPU"):
+                continue
+
+            try:
+                pci = core.get_property(device, "DEVICE_PCI_INFO")
+                raw_name = core.get_property(device, "FULL_DEVICE_NAME")
+                device_type = core.get_property(device, "DEVICE_TYPE")
+            except Exception as exc:
+                logger.debug(f"Failed to read properties for {device}: {exc}")
+                continue
+
+            pdev = self._format_pdev(pci)
+            if not pdev:
+                continue
+
+            names[pdev] = self._resolve_name(raw_name, device_type, cpu_name)
+
+        self._names = names
+        return names
+
+    @staticmethod
+    def _format_pdev(pci) -> Optional[str]:
+        try:
+            return f"{pci.domain:04x}:{pci.bus:02x}:{pci.device:02x}.{pci.function:x}"
+        except AttributeError:
+            return None
+
+    @classmethod
+    def _resolve_name(cls, raw_name: str, device_type, cpu_name: Optional[str]) -> str:
+        """Build a display name for a GPU.
+
+        Modern integrated Intel GPUs are reported by OpenVINO with a generic
+        FULL_DEVICE_NAME like "Intel(R) Graphics (iGPU)" that gives no model
+        information. Since the iGPU is part of the CPU on these platforms, fall
+        back to the CPU name (which OpenVINO does report specifically) and
+        suffix it with "iGPU" so it's clear what the entry is.
+        """
+        is_integrated = "INTEGRATED" in str(device_type).upper()
+
+        if is_integrated and cpu_name:
+            short_cpu = re.sub(r"^Intel\s+", "", cpu_name)
+            return f"{short_cpu} iGPU"
+
+        return cls._normalize_name(raw_name)
+
+    @classmethod
+    def _normalize_name(cls, name: str) -> str:
+        cleaned = cls._strip_trademarks(name)
+        cleaned = re.sub(r"\s*\((?:i|d)GPU\)\s*$", "", cleaned, flags=re.IGNORECASE)
+        return " ".join(cleaned.split())
+
+    @staticmethod
+    def _strip_trademarks(name: str) -> str:
+        cleaned = re.sub(r"\(R\)|\(TM\)", "", name)
+        return " ".join(cleaned.split())
+
+
+intel_gpu_name_resolver = IntelGpuNameResolver()

frigate/stats/util.py

@@ -230,6 +230,7 @@ async def set_gpu_stats(
                 hwaccel_args.append(args)
 
     stats: dict[str, dict] = {}
+    intel_gpu_collected = False
 
     for args in hwaccel_args:
         if args in hwaccel_errors:
@@ -242,6 +243,7 @@ async def set_gpu_stats(
             if nvidia_usage:
                 for i in range(len(nvidia_usage)):
                     stats[nvidia_usage[i]["name"]] = {
+                        "vendor": "nvidia",
                         "gpu": str(round(float(nvidia_usage[i]["gpu"]), 2)) + "%",
                         "mem": str(round(float(nvidia_usage[i]["mem"]), 2)) + "%",
                         "enc": str(round(float(nvidia_usage[i]["enc"]), 2)) + "%",
@@ -250,31 +252,34 @@ async def set_gpu_stats(
                     }
 
             else:
-                stats["nvidia-gpu"] = {"gpu": "", "mem": ""}
+                stats["nvidia-gpu"] = {"vendor": "nvidia", "gpu": "", "mem": ""}
                 hwaccel_errors.append(args)
         elif "nvmpi" in args or "jetson" in args:
             # nvidia Jetson
             jetson_usage = get_jetson_stats()
 
             if jetson_usage:
-                stats["jetson-gpu"] = jetson_usage
+                stats["jetson-gpu"] = {"vendor": "nvidia", **jetson_usage}
             else:
-                stats["jetson-gpu"] = {"gpu": "", "mem": ""}
+                stats["jetson-gpu"] = {"vendor": "nvidia", "gpu": "", "mem": ""}
                 hwaccel_errors.append(args)
         elif "qsv" in args or ("vaapi" in args and not is_vaapi_amd_driver()):
             if not config.telemetry.stats.intel_gpu_stats:
                 continue
 
-            if "intel-gpu" not in stats:
+            if not intel_gpu_collected:
                 # intel GPU (QSV or VAAPI both use the same physical GPU)
+                intel_gpu_collected = True
                 intel_usage = get_intel_gpu_stats(
                     config.telemetry.stats.intel_gpu_device
                 )
 
-                if intel_usage is not None:
-                    stats["intel-gpu"] = intel_usage or {"gpu": "", "mem": ""}
+                if intel_usage:
+                    for entry in intel_usage.values():
+                        name = entry.pop("name")
+                        stats[name] = entry
                 else:
-                    stats["intel-gpu"] = {"gpu": "", "mem": ""}
+                    stats["intel-gpu"] = {"vendor": "intel", "gpu": "", "mem": ""}
                     hwaccel_errors.append(args)
         elif "vaapi" in args:
             if not config.telemetry.stats.amd_gpu_stats:
@@ -284,18 +289,18 @@ async def set_gpu_stats(
             amd_usage = get_amd_gpu_stats()
 
             if amd_usage:
-                stats["amd-vaapi"] = amd_usage
+                stats["amd-vaapi"] = {"vendor": "amd", **amd_usage}
             else:
-                stats["amd-vaapi"] = {"gpu": "", "mem": ""}
+                stats["amd-vaapi"] = {"vendor": "amd", "gpu": "", "mem": ""}
                 hwaccel_errors.append(args)
         elif "preset-rk" in args:
             rga_usage = get_rockchip_gpu_stats()
 
             if rga_usage:
-                stats["rockchip"] = rga_usage
+                stats["rockchip"] = {"vendor": "rockchip", **rga_usage}
         elif "v4l2m2m" in args or "rpi" in args:
             # RPi v4l2m2m is currently not able to get usage stats
-            stats["rpi-v4l2m2m"] = {"gpu": "", "mem": ""}
+            stats["rpi-v4l2m2m"] = {"vendor": "rpi", "gpu": "", "mem": ""}
 
     if stats:
         all_stats["gpu_usages"] = stats

frigate/test/http_api/test_debug_replay_api.py

@@ -0,0 +1,123 @@
+"""Tests for /debug_replay API endpoints."""
+
+from unittest.mock import patch
+
+from frigate.models import Event, Recordings, ReviewSegment
+from frigate.test.http_api.base_http_test import AuthTestClient, BaseTestHttp
+
+
+class TestDebugReplayAPI(BaseTestHttp):
+    def setUp(self):
+        super().setUp([Event, Recordings, ReviewSegment])
+        self.app = self.create_app()
+
+    def test_start_returns_202_with_job_id(self):
+        # Stub the factory to skip validation/threading and just record the
+        # name on the manager the way the real factory's mark_starting would.
+        def fake_start(**kwargs):
+            kwargs["replay_manager"].mark_starting(
+                source_camera=kwargs["source_camera"],
+                replay_camera_name="_replay_front",
+                start_ts=kwargs["start_ts"],
+                end_ts=kwargs["end_ts"],
+            )
+            return "job-1234"
+
+        with patch(
+            "frigate.api.debug_replay.start_debug_replay_job",
+            side_effect=fake_start,
+        ):
+            with AuthTestClient(self.app) as client:
+                resp = client.post(
+                    "/debug_replay/start",
+                    json={
+                        "camera": "front",
+                        "start_time": 100,
+                        "end_time": 200,
+                    },
+                )
+
+        self.assertEqual(resp.status_code, 202)
+        body = resp.json()
+        self.assertTrue(body["success"])
+        self.assertEqual(body["job_id"], "job-1234")
+        self.assertEqual(body["replay_camera"], "_replay_front")
+
+    def test_start_returns_400_on_validation_error(self):
+        with patch(
+            "frigate.api.debug_replay.start_debug_replay_job",
+            side_effect=ValueError("Camera 'missing' not found"),
+        ):
+            with AuthTestClient(self.app) as client:
+                resp = client.post(
+                    "/debug_replay/start",
+                    json={
+                        "camera": "missing",
+                        "start_time": 100,
+                        "end_time": 200,
+                    },
+                )
+
+        self.assertEqual(resp.status_code, 400)
+        body = resp.json()
+        self.assertFalse(body["success"])
+        # Message is hard-coded so we don't echo exception text back to clients
+        # (CodeQL: information exposure through an exception).
+        self.assertEqual(body["message"], "Invalid debug replay parameters")
+
+    def test_start_returns_409_when_session_already_active(self):
+        with patch(
+            "frigate.api.debug_replay.start_debug_replay_job",
+            side_effect=RuntimeError("A replay session is already active"),
+        ):
+            with AuthTestClient(self.app) as client:
+                resp = client.post(
+                    "/debug_replay/start",
+                    json={
+                        "camera": "front",
+                        "start_time": 100,
+                        "end_time": 200,
+                    },
+                )
+
+        self.assertEqual(resp.status_code, 409)
+        body = resp.json()
+        self.assertFalse(body["success"])
+
+    def test_status_inactive_when_no_session(self):
+        with AuthTestClient(self.app) as client:
+            resp = client.get("/debug_replay/status")
+
+        self.assertEqual(resp.status_code, 200)
+        body = resp.json()
+        self.assertFalse(body["active"])
+        self.assertIsNone(body["replay_camera"])
+        self.assertIsNone(body["source_camera"])
+        self.assertIsNone(body["start_time"])
+        self.assertIsNone(body["end_time"])
+        self.assertFalse(body["live_ready"])
+        # Make sure deprecated fields are gone
+        self.assertNotIn("state", body)
+        self.assertNotIn("progress_percent", body)
+        self.assertNotIn("error_message", body)
+
+    def test_status_active_after_mark_starting(self):
+        manager = self.app.replay_manager
+        manager.mark_starting(
+            source_camera="front",
+            replay_camera_name="_replay_front",
+            start_ts=100.0,
+            end_ts=200.0,
+        )
+
+        with AuthTestClient(self.app) as client:
+            resp = client.get("/debug_replay/status")
+
+        self.assertEqual(resp.status_code, 200)
+        body = resp.json()
+        self.assertTrue(body["active"])
+        self.assertEqual(body["replay_camera"], "_replay_front")
+        self.assertEqual(body["source_camera"], "front")
+        self.assertEqual(body["start_time"], 100.0)
+        self.assertEqual(body["end_time"], 200.0)
+        self.assertFalse(body["live_ready"])

frigate/test/http_api/test_http_app.py

@@ -23,6 +23,26 @@ class TestHttpApp(BaseTestHttp):
             response_json = response.json()
             assert response_json == self.test_stats
 
+    def test_recordings_storage_requires_admin(self):
+        stats = Mock(spec=StatsEmitter)
+        stats.get_latest_stats.return_value = self.test_stats
+        app = super().create_app(stats)
+        app.storage_maintainer = Mock()
+        app.storage_maintainer.calculate_camera_usages.return_value = {
+            "front_door": {"usage": 2.0},
+        }
+
+        with AuthTestClient(app) as client:
+            response = client.get(
+                "/recordings/storage",
+                headers={"remote-user": "viewer", "remote-role": "viewer"},
+            )
+            assert response.status_code == 403
+
+            response = client.get("/recordings/storage")
+            assert response.status_code == 200
+            assert response.json()["front_door"]["usage_percent"] == 25.0
+
     def test_config_set_in_memory_replaces_objects_track_list(self):
         self.minimal_config["cameras"]["front_door"]["objects"] = {
             "track": ["person", "car"],

frigate/test/http_api/test_http_camera_access.py

@@ -1,3 +1,4 @@
+import os
 from unittest.mock import patch
 
 from fastapi import HTTPException, Request
@@ -357,6 +358,51 @@ class TestGo2rtcStreamAccess(BaseTestHttp):
             f"got {resp.status_code}"
         )
 
+    def test_add_stream_rejects_restricted_source(self):
+        """PUT /go2rtc/streams must reject exec:/echo:/expr: sources even for
+        admins"""
+        app = self._make_app(_MULTI_CAMERA_CONFIG)
+        with AuthTestClient(app) as client:
+            for src in (
+                "exec:/tmp/rev.sh",
+                "echo:foo",
+                "expr:bar",
+                "  exec:/tmp/rev.sh",
+            ):
+                resp = client.put(f"/go2rtc/streams/revshell?src={src}")
+                assert resp.status_code == 400, (
+                    f"Expected 400 for restricted src {src!r}; got {resp.status_code}"
+                )
+                assert resp.json().get("success") is False
+
+    def test_add_stream_allows_non_restricted_source(self):
+        """A normal stream URL should pass the restricted-source check and reach
+        the (unavailable in tests) go2rtc proxy — so we expect 500, not 400."""
+        app = self._make_app(_MULTI_CAMERA_CONFIG)
+        with AuthTestClient(app) as client:
+            resp = client.put("/go2rtc/streams/legit?src=rtsp://10.0.0.1:554/video")
+            assert resp.status_code != 400, (
+                f"Non-restricted source should not be rejected with 400; got {resp.status_code}"
+            )
+
+    def test_add_stream_allows_restricted_source_when_override_set(self):
+        """When GO2RTC_ALLOW_ARBITRARY_EXEC is set, the API must defer to operator
+        intent and forward the request to go2rtc instead of short-circuiting with 400."""
+        app = self._make_app(_MULTI_CAMERA_CONFIG)
+        mock_response = type("R", (), {"ok": True, "status_code": 200, "text": "ok"})()
+        with patch.dict(os.environ, {"GO2RTC_ALLOW_ARBITRARY_EXEC": "true"}):
+            with patch(
+                "frigate.api.camera.requests.put", return_value=mock_response
+            ) as mock_put:
+                with AuthTestClient(app) as client:
+                    resp = client.put("/go2rtc/streams/legit?src=exec:/tmp/something")
+                assert resp.status_code == 200, (
+                    f"Restricted src should be forwarded when override set; got {resp.status_code}"
+                )
+                mock_put.assert_called_once()
+                forwarded_src = mock_put.call_args.kwargs["params"]["src"]
+                assert forwarded_src == "exec:/tmp/something"
+
     def test_stream_alias_blocked_when_owning_camera_disallowed(self):
         """limited_user cannot access a stream alias that belongs to a camera they
         are not allowed to see."""

frigate/test/http_api/test_http_event.py

@@ -219,6 +219,25 @@ class TestHttpApp(BaseTestHttp):
             assert len(events) == 1
             assert events[0]["id"] == event_id
 
+    def test_similarity_search_hides_unauthorized_anchor_event(self):
+        mock_embeddings = Mock()
+        self.app.frigate_config.semantic_search.enabled = True
+        self.app.embeddings = mock_embeddings
+
+        with AuthTestClient(self.app) as client:
+            super().insert_mock_event("hidden.anchor", camera="back_door")
+            response = client.get(
+                "/events/search",
+                params={
+                    "search_type": "similarity",
+                    "event_id": "hidden.anchor",
+                },
+            )
+
+        assert response.status_code == 404
+        assert response.json()["message"] == "Event not found"
+        mock_embeddings.search_thumbnail.assert_not_called()
+
     def test_get_good_event(self):
         id = "123456.random"
 

frigate/test/test_chat_find_similar_objects.py

@@ -145,9 +145,12 @@ class TestExecuteFindSimilarObjects(unittest.TestCase):
             embeddings=embeddings,
             frigate_config=SimpleNamespace(
                 semantic_search=SimpleNamespace(enabled=semantic_enabled),
+                cameras={"driveway": object()},
+                auth=SimpleNamespace(roles={"admin": [], "viewer": ["driveway"]}),
+                proxy=SimpleNamespace(separator=","),
             ),
         )
-        return SimpleNamespace(app=app)
+        return SimpleNamespace(app=app, headers={})
 
     def test_semantic_search_disabled_returns_error(self):
         req = self._make_request(semantic_enabled=False)
@@ -180,7 +183,7 @@ class TestExecuteFindSimilarObjects(unittest.TestCase):
             _execute_find_similar_objects(
                 req,
                 {"event_id": "anchor", "cameras": ["nonexistent_cam"]},
-                allowed_cameras=["nonexistent_cam"],
+                allowed_cameras=["driveway"],
             )
         )
         self.assertEqual(result["results"], [])

frigate/test/test_config.py

@@ -10,7 +10,7 @@ from ruamel.yaml.constructor import DuplicateKeyError
 from frigate.config import BirdseyeModeEnum, FrigateConfig
 from frigate.const import MODEL_CACHE_DIR
 from frigate.detectors import DetectorTypeEnum
-from frigate.util.builtin import deep_merge, load_labels
+from frigate.util.builtin import deep_merge
 
 
 class TestConfig(unittest.TestCase):
@@ -64,9 +64,9 @@ class TestConfig(unittest.TestCase):
 
     def test_config_class(self):
         frigate_config = FrigateConfig(**self.minimal)
-        assert "cpu" in frigate_config.detectors.keys()
-        assert frigate_config.detectors["cpu"].type == DetectorTypeEnum.cpu
-        assert frigate_config.detectors["cpu"].model.width == 320
+        assert "ov" in frigate_config.detectors.keys()
+        assert frigate_config.detectors["ov"].type == DetectorTypeEnum.openvino
+        assert frigate_config.detectors["ov"].model.width == 300
 
     @patch("frigate.detectors.detector_config.load_labels")
     def test_detector_custom_model_path(self, mock_labels):
@@ -309,16 +309,11 @@ class TestConfig(unittest.TestCase):
         }
 
         frigate_config = FrigateConfig(**config)
-        all_audio_labels = {
-            label
-            for label in load_labels("/audio-labelmap.txt", prefill=521).values()
-            if label
+        assert set(frigate_config.cameras["back"].audio.filters.keys()) == {
+            "speech",
+            "yell",
         }
 
-        assert all_audio_labels.issubset(
-            set(frigate_config.cameras["back"].audio.filters.keys())
-        )
-
     def test_override_audio_filters(self):
         config = {
             "mqtt": {"host": "mqtt"},
@@ -345,7 +340,8 @@ class TestConfig(unittest.TestCase):
         frigate_config = FrigateConfig(**config)
         assert "speech" in frigate_config.cameras["back"].audio.filters
         assert frigate_config.cameras["back"].audio.filters["speech"].threshold == 0.9
-        assert "babbling" in frigate_config.cameras["back"].audio.filters
+        assert "yell" in frigate_config.cameras["back"].audio.filters
+        assert "babbling" not in frigate_config.cameras["back"].audio.filters
 
     def test_inherit_object_filters(self):
         config = {
@@ -1005,6 +1001,7 @@ class TestConfig(unittest.TestCase):
 
         config = {
             "mqtt": {"host": "mqtt"},
+            "detectors": {"cpu": {"type": "cpu"}},
             "model": {"path": "plus://test"},
             "cameras": {
                 "back": {

frigate/test/test_debug_replay.py

@@ -0,0 +1,242 @@
+"""Tests for the simplified DebugReplayManager.
+
+Startup orchestration lives in ``frigate.jobs.debug_replay`` (covered by
+``test_debug_replay_job``). The manager owns only session presence and
+cleanup.
+"""
+
+import unittest
+import unittest.mock
+from unittest.mock import MagicMock, patch
+
+
+class TestDebugReplayManagerSession(unittest.TestCase):
+    def test_inactive_by_default(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+
+        self.assertFalse(manager.active)
+        self.assertIsNone(manager.replay_camera_name)
+        self.assertIsNone(manager.source_camera)
+        self.assertIsNone(manager.clip_path)
+        self.assertIsNone(manager.start_ts)
+        self.assertIsNone(manager.end_ts)
+
+    def test_mark_starting_sets_session_pointers_and_active(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+
+        manager.mark_starting(
+            source_camera="front",
+            replay_camera_name="_replay_front",
+            start_ts=100.0,
+            end_ts=200.0,
+        )
+
+        self.assertTrue(manager.active)
+        self.assertEqual(manager.replay_camera_name, "_replay_front")
+        self.assertEqual(manager.source_camera, "front")
+        self.assertEqual(manager.start_ts, 100.0)
+        self.assertEqual(manager.end_ts, 200.0)
+        self.assertIsNone(manager.clip_path)
+
+    def test_mark_session_ready_sets_clip_path(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        manager.mark_starting("front", "_replay_front", 100.0, 200.0)
+
+        manager.mark_session_ready(clip_path="/tmp/replay/_replay_front.mp4")
+
+        self.assertEqual(manager.clip_path, "/tmp/replay/_replay_front.mp4")
+        self.assertTrue(manager.active)
+
+    def test_clear_session_resets_all_pointers(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        manager.mark_starting("front", "_replay_front", 100.0, 200.0)
+        manager.mark_session_ready("/tmp/replay/clip.mp4")
+
+        manager.clear_session()
+
+        self.assertFalse(manager.active)
+        self.assertIsNone(manager.replay_camera_name)
+        self.assertIsNone(manager.source_camera)
+        self.assertIsNone(manager.clip_path)
+        self.assertIsNone(manager.start_ts)
+        self.assertIsNone(manager.end_ts)
+
+
+class TestDebugReplayManagerStop(unittest.TestCase):
+    def test_stop_when_inactive_is_a_noop(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        frigate_config = MagicMock()
+        frigate_config.cameras = {}
+        publisher = MagicMock()
+
+        # Should not raise; should not publish any events.
+        manager.stop(frigate_config=frigate_config, config_publisher=publisher)
+
+        publisher.publish_update.assert_not_called()
+
+    def test_stop_publishes_remove_when_camera_was_published(self) -> None:
+        from frigate.config.camera.updater import CameraConfigUpdateEnum
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        manager.mark_starting("front", "_replay_front", 100.0, 200.0)
+        manager.mark_session_ready("/tmp/replay/_replay_front.mp4")
+
+        camera_config = MagicMock()
+        frigate_config = MagicMock()
+        frigate_config.cameras = {"_replay_front": camera_config}
+        publisher = MagicMock()
+
+        with (
+            patch.object(manager, "_cleanup_db"),
+            patch.object(manager, "_cleanup_files"),
+            patch("frigate.debug_replay.cancel_debug_replay_job", return_value=False),
+        ):
+            manager.stop(frigate_config=frigate_config, config_publisher=publisher)
+
+        # One publish_update call with a remove topic.
+        self.assertEqual(publisher.publish_update.call_count, 1)
+        topic_arg = publisher.publish_update.call_args.args[0]
+        self.assertEqual(topic_arg.update_type, CameraConfigUpdateEnum.remove)
+        self.assertFalse(manager.active)
+
+    def test_stop_skips_remove_publish_when_camera_not_in_config(self) -> None:
+        """Cancellation during preparing_clip: no camera was published yet."""
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        manager.mark_starting("front", "_replay_front", 100.0, 200.0)
+        # clip_path stays None because we cancelled before camera publish.
+
+        frigate_config = MagicMock()
+        frigate_config.cameras = {}  # _replay_front not present
+        publisher = MagicMock()
+
+        with (
+            patch.object(manager, "_cleanup_db"),
+            patch.object(manager, "_cleanup_files"),
+            patch("frigate.debug_replay.cancel_debug_replay_job", return_value=True),
+        ):
+            manager.stop(frigate_config=frigate_config, config_publisher=publisher)
+
+        publisher.publish_update.assert_not_called()
+        self.assertFalse(manager.active)
+
+    def test_stop_calls_cancel_debug_replay_job(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        manager.mark_starting("front", "_replay_front", 100.0, 200.0)
+
+        frigate_config = MagicMock()
+        frigate_config.cameras = {}
+        publisher = MagicMock()
+
+        with (
+            patch.object(manager, "_cleanup_db"),
+            patch.object(manager, "_cleanup_files"),
+            patch(
+                "frigate.debug_replay.cancel_debug_replay_job",
+                return_value=True,
+            ) as mock_cancel,
+        ):
+            manager.stop(frigate_config=frigate_config, config_publisher=publisher)
+
+        mock_cancel.assert_called_once()
+
+
+class TestDebugReplayManagerPublishCamera(unittest.TestCase):
+    def test_publish_camera_invokes_publisher_with_add_topic(self) -> None:
+        from frigate.config.camera.updater import CameraConfigUpdateEnum
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+
+        source_config = MagicMock()
+        new_camera_config = MagicMock()
+        frigate_config = MagicMock()
+        frigate_config.cameras = {"front": source_config}
+        publisher = MagicMock()
+
+        with (
+            patch.object(
+                manager,
+                "_build_camera_config_dict",
+                return_value={"enabled": True},
+            ),
+            patch("frigate.debug_replay.find_config_file", return_value="/cfg.yml"),
+            patch("frigate.debug_replay.YAML") as yaml_cls,
+            patch("frigate.debug_replay.FrigateConfig.parse_object") as parse_object,
+            patch("builtins.open", unittest.mock.mock_open(read_data="cameras:\n")),
+        ):
+            yaml_instance = yaml_cls.return_value
+            yaml_instance.load.return_value = {"cameras": {}}
+            parsed = MagicMock()
+            parsed.cameras = {"_replay_front": new_camera_config}
+            parse_object.return_value = parsed
+
+            manager.publish_camera(
+                source_camera="front",
+                replay_name="_replay_front",
+                clip_path="/tmp/clip.mp4",
+                frigate_config=frigate_config,
+                config_publisher=publisher,
+            )
+
+        # Camera registered into the live config dict
+        self.assertIn("_replay_front", frigate_config.cameras)
+        # Publisher invoked with an add topic
+        self.assertEqual(publisher.publish_update.call_count, 1)
+        topic_arg = publisher.publish_update.call_args.args[0]
+        self.assertEqual(topic_arg.update_type, CameraConfigUpdateEnum.add)
+
+    def test_publish_camera_wraps_parse_failure_in_runtime_error(self) -> None:
+        from frigate.debug_replay import DebugReplayManager
+
+        manager = DebugReplayManager()
+        frigate_config = MagicMock()
+        frigate_config.cameras = {"front": MagicMock()}
+        publisher = MagicMock()
+
+        with (
+            patch.object(
+                manager,
+                "_build_camera_config_dict",
+                return_value={"enabled": True},
+            ),
+            patch("frigate.debug_replay.find_config_file", return_value="/cfg.yml"),
+            patch("frigate.debug_replay.YAML") as yaml_cls,
+            patch(
+                "frigate.debug_replay.FrigateConfig.parse_object",
+                side_effect=ValueError("zone foo has invalid coordinates"),
+            ),
+            patch("builtins.open", unittest.mock.mock_open(read_data="cameras:\n")),
+        ):
+            yaml_cls.return_value.load.return_value = {"cameras": {}}
+
+            with self.assertRaises(RuntimeError) as ctx:
+                manager.publish_camera(
+                    source_camera="front",
+                    replay_name="_replay_front",
+                    clip_path="/tmp/clip.mp4",
+                    frigate_config=frigate_config,
+                    config_publisher=publisher,
+                )
+
+        self.assertIn("replay camera config", str(ctx.exception))
+        self.assertIn("invalid coordinates", str(ctx.exception))
+        publisher.publish_update.assert_not_called()
+
+
+if __name__ == "__main__":
+    unittest.main()