Fix chat tool calling and prompt breaking (#23457)

* Implement tool call history keeping

* Refactor to match single message implementation

* Simplify data representation

* Cleanup chat page rendering

* Include system message to not break cache

* Formatting

* Update tests and update .gitignore

docs/docs/frigate/glossary.md

@@ -5,20 +5,40 @@ title: Glossary
 
 The glossary explains terms commonly used in Frigate's documentation.
 
+## Alert
+
+The higher-priority of the two [review item](#review-item) severities, the other being a [detection](#detection). By default a review item is an alert when it involves a `person` or `car`; the qualifying [labels](#label) and [zones](#zone) can be configured. [See the review docs for more info](/configuration/review)
+
+## Attribute
+
+A property detected on an [object](#object) that exists alongside its [label](#label). Unlike a [sub label](#sub-label), an object can carry several attributes at once. Some attributes come directly from the object detection [model](#model) — for example `face`, `license_plate`, or delivery carrier logos such as `amazon`, `ups`, and `fedex` — while others come from a [custom object classification model](/configuration/custom_classification/object_classification) configured with the `attribute` type. Attributes are visible in the Tracked Object Details pane in Explore, in `frigate/events` MQTT messages, and through the HTTP API.
+
 ## Bounding Box
 
-A box returned from the object detection model that outlines an object in the frame. These have multiple colors depending on object type in the debug live view.
+A box returned by the object detection [model](#model) that outlines a detected [object](#object) in the frame. In the Debug view, bounding boxes are colored by object [label](#label).
 
 ### Bounding Box Colors
 
 - At startup different colors will be assigned to each object label
 - A dark blue thin line indicates that object is not detected at this current point in time
 - A gray thin line indicates that object is detected as being stationary
-- A thick line indicates that object is the subject of autotracking (when enabled).
+- A thick line indicates that object is the subject of autotracking (when enabled)
+
+## Class
+
+The categories a classification [model](#model) is trained to distinguish between. Each class is a distinct visual category the model predicts, plus a `none` class for inputs that don't fit any category. For example, a custom object classification model for `person` objects might use the classes `delivery_person`, `resident`, and `none`. The predicted class is applied to the [object](#object) as either a [sub label](#sub-label) or an [attribute](#attribute), depending on the model's configuration. [See the object classification docs for more info](/configuration/custom_classification/object_classification)
+
+## Detection
+
+The lower-priority of the two [review item](#review-item) severities, the other being an [alert](#alert). By default, any review item that does not qualify as an alert is a detection; the qualifying [labels](#label) and [zones](#zone) can be configured. Despite the name, a detection is a category of review item — not the same as the object detection performed by the [model](#model). [See the review docs for more info](/configuration/review)
 
 ## False Positive
 
-An incorrect detection of an object type. For example a dog being detected as a person, a chair being detected as a dog, etc. A person being detected in an area you want to ignore is not a false positive.
+An incorrect result from the object detection [model](#model), where it assigns the wrong [label](#label) to something in the frame — for example a dog identified as a person, or a chair identified as a dog. A person correctly identified in an area you want to ignore is not a false positive.
+
+## Label
+
+The type assigned to a detected [object](#object) by the object detection [model](#model), drawn from the model's labelmap — for example `person`, `car`, or `dog`. Frigate tracks `person` by default; additional labels are tracked by adding them to the objects configuration. [See the available objects docs for the full list](/configuration/objects)
 
 ## Mask
 
@@ -26,44 +46,56 @@ There are two types of masks in Frigate. [See the mask docs for more info](/conf
 
 ### Motion Mask
 
-Motion masks prevent detection of [motion](#motion) in masked areas from triggering Frigate to run object detection, but do not prevent objects from being detected if object detection runs due to motion in nearby areas. For example: camera timestamps, skies, the tops of trees, etc.
+A motion mask stops [motion](#motion) in the masked area from triggering object detection. It does not stop an object from being detected when object detection runs because of motion in a nearby area. Use motion masks for parts of the frame that change constantly but never contain objects you care about — camera timestamps, the sky, the tops of trees, and so on.
 
 ### Object Mask
 
-Object filter masks drop any bounding boxes where the bottom center (overlap doesn't matter) is in the masked area. It forces them to be considered a [false positive](#false-positive) so that they are ignored.
+An object filter mask drops any [bounding box](#bounding-box) whose bottom center falls inside the masked area (overlap elsewhere doesn't matter). The object is forced to be treated as a [false positive](#false-positive) and ignored.
 
 ## Min Score
 
-The lowest score that an object can be detected with during tracking, any detection with a lower score will be assumed to be a false positive
+The lowest score a detected object can have to be kept during tracking. Anything scoring below the minimum is assumed to be a [false positive](#false-positive) and discarded.
+
+## Model
+
+A machine learning model that Frigate uses to detect or classify objects. The object detection model locates [objects](#object) in each frame and returns their [labels](#label) and [bounding boxes](#bounding-box). Additional enrichment models run on tracked objects to add detail: face recognition, license plate recognition, bird classification, custom object and state classification, and the embedding models used for semantic search. [See the object detectors docs for more info](/configuration/object_detectors)
 
 ## Motion
 
-When pixels in the current camera frame are different than previous frames. When many nearby pixels are different in the current frame they grouped together and indicated with a red motion box in the live debug view. [See the motion detection docs for more info](/configuration/motion_detection)
+A change in pixels between the current camera frame and previous frames. When many nearby pixels change together, they are grouped and shown as a red motion box in the debug live view. [See the motion detection docs for more info](/configuration/motion_detection)
+
+## Object
+
+Something Frigate can detect and follow in a camera frame, identified by its [label](#label) (for example a person or a car). The object types Frigate watches for are set in the `objects` configuration. Once an object is detected and followed across frames it becomes a [tracked object](#tracked-object-event-in-previous-versions), which may also carry a [sub label](#sub-label) and [attributes](#attribute). [See the available objects docs for more info](/configuration/objects)
 
 ## Region
 
-A portion of the camera frame that is sent to object detection, regions can be sent due to motion, active objects, or occasionally for stationary objects. These are represented by green boxes in the debug live view.
+A portion of the camera frame sent to the object detection [model](#model). Regions are selected because of [motion](#motion), active objects, or occasionally to recheck stationary objects, and are shown as green boxes in the debug live view.
 
 ## Review Item
 
-A review item is a time period where any number of events/tracked objects were active. [See the review docs for more info](/configuration/review)
+A period of time during which one or more [tracked objects](#tracked-object-event-in-previous-versions) were active, grouped together for review. Each review item is categorized as either an [alert](#alert) or a [detection](#detection). [See the review docs for more info](/configuration/review)
 
 ## Snapshot Score
 
-The score shown in a snapshot is the score of that object at that specific moment in time.
+The object's score at the specific moment the snapshot was captured.
+
+## Sub Label
+
+A more specific identity assigned to a [tracked object](#tracked-object-event-in-previous-versions) in addition to its [label](#label). A `person` may get the name of a recognized face, a `car` may get the name of a known license plate, and a `bird` may get its species. An object can have only one sub label at a time. Sub labels are produced by face recognition, license plate recognition, bird classification, custom object classification configured with the `sub label` type, and semantic search triggers.
 
 ## Threshold
 
-The threshold is the median score that an object must reach in order to be considered a true positive.
+The median score an object must reach to be considered a true positive.
 
 ## Top Score
 
-The top score for an object is the highest median score for an object.
+The highest median score an object reached over its lifetime.
 
 ## Tracked Object ("event" in previous versions)
 
-The time period starting when a tracked object entered the frame and ending when it left the frame, including any time that the object remained still. Tracked objects are saved when it is considered a [true positive](#threshold) and meets the requirements for a snapshot or recording to be saved.
+An [object](#object) followed from the moment it enters the frame until it leaves, including any time it stays still. A tracked object is saved once it is considered a [true positive](#threshold) and meets the requirements for a snapshot or recording.
 
 ## Zone
 
-Zones are areas of interest, zones can be used for notifications and for limiting the areas where Frigate will create a [review item](#review-item). [See the zone docs for more info](/configuration/zones)
+A user-defined area of interest within the camera frame. Zones can be used for notifications and to limit where Frigate creates a [review item](#review-item). [See the zone docs for more info](/configuration/zones)

docs/docs/troubleshooting/recordings.md

@@ -121,6 +121,12 @@ If segments are only ~1 second instead of ~10 seconds, the camera is sending cor
 - **Changing codec, bitrate, or resolution mid-stream** — Any encoding changes during an active stream can cause unpredictable segment splitting.
 - **Camera firmware bugs** — Check for firmware updates from your camera manufacturer.
 
+:::tip
+
+You don't have to run `ffprobe` by hand to catch this. Open a camera's **Camera Probe Info** dialog (the info icon on the System → Metrics → Cameras page) and check the **Keyframe analysis** section. It probes the record stream and flags sparse or variable keyframes, which is what smart/"+" codecs (H.264+/H.265+) and long keyframe intervals produce.
+
+:::
+
 ### Step 4: Check for a stuck detector
 
 If the detect stream is not processing frames, segments will accumulate. Common causes:

docs/static/frigate-api.yaml

@@ -400,6 +400,35 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/HTTPValidationError"
+  /keyframe_analysis:
+    get:
+      tags:
+        - Camera
+      summary: Keyframe Analysis
+      description: >-
+        Probe a camera's record stream and classify its keyframe spacing.
+        Detects smart/+ codecs and long/variable GOPs that degrade recording.
+      operationId: keyframe_analysis_keyframe_analysis_get
+      parameters:
+        - name: camera
+          in: query
+          required: false
+          schema:
+            type: string
+            default: ""
+            title: Camera
+      responses:
+        "200":
+          description: Successful Response
+          content:
+            application/json:
+              schema: {}
+        "422":
+          description: Validation Error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/HTTPValidationError"
   /ffprobe/snapshot:
     get:
       tags:

frigate/api/camera.py

@@ -34,11 +34,15 @@ from frigate.config.camera.updater import (
 )
 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.builtin import clean_camera_user_pass, get_record_segment_time
 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, is_restricted_go2rtc_source
+from frigate.util.services import (
+    analyze_record_keyframes,
+    ffprobe_stream,
+    is_restricted_go2rtc_source,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -362,6 +366,48 @@ def ffprobe(request: Request, paths: str = "", detailed: bool = False):
     return JSONResponse(content=output)
 
 
+@router.get("/keyframe_analysis", dependencies=[Depends(require_role(["admin"]))])
+async def keyframe_analysis(request: Request, camera: str = ""):
+    """Probe a camera's record stream and classify its keyframe spacing.
+
+    Detects smart/+ codecs and long/variable GOPs that degrade recording.
+    """
+    config: FrigateConfig = request.app.frigate_config
+
+    if camera not in config.cameras:
+        return JSONResponse(
+            content={"success": False, "message": f"{camera} is not a valid camera."},
+            status_code=404,
+        )
+
+    camera_config = config.cameras[camera]
+
+    if not camera_config.enabled:
+        return JSONResponse(
+            content={"success": False, "message": f"{camera} is not enabled."},
+            status_code=404,
+        )
+
+    # keyframe spacing only matters when this camera is recording
+    if not camera_config.record.enabled:
+        return JSONResponse(content={"severity": "record_disabled"})
+
+    # recording guarantees an input carries the record role; its index matches
+    # the "Stream N" numbering the ffprobe endpoint surfaces (same input order)
+    record_index, record_input = next(
+        (idx, i)
+        for idx, i in enumerate(camera_config.ffmpeg.inputs)
+        if "record" in i.roles
+    )
+
+    segment_time = get_record_segment_time(camera_config)
+    result = await analyze_record_keyframes(
+        config.ffmpeg, record_input.path, segment_time
+    )
+    result["stream_index"] = record_index
+    return JSONResponse(content=result)
+
+
 @router.get("/ffprobe/snapshot", dependencies=[Depends(require_role(["admin"]))])
 def ffprobe_snapshot(request: Request, url: str = "", timeout: int = 10):
     """Get a snapshot from a stream URL using ffmpeg."""

frigate/api/chat.py

@@ -7,7 +7,7 @@ import operator
 import time
 from datetime import datetime
 from functools import reduce
-from typing import Any, Dict, List, Optional
+from typing import Any, Optional
 
 import cv2
 from fastapi import APIRouter, Body, Depends, HTTPException, Request
@@ -59,7 +59,7 @@ class ToolExecuteRequest(BaseModel):
     """Request model for tool execution."""
 
     tool_name: str
-    arguments: Dict[str, Any]
+    arguments: dict[str, Any]
 
 
 class VLMMonitorRequest(BaseModel):
@@ -68,8 +68,8 @@ class VLMMonitorRequest(BaseModel):
     camera: str
     condition: str
     max_duration_minutes: int = 60
-    labels: List[str] = []
-    zones: List[str] = []
+    labels: list[str] = []
+    zones: list[str] = []
 
 
 @router.get(
@@ -91,10 +91,10 @@ def get_tools(request: Request) -> JSONResponse:
 
 
 def _resolve_zones(
-    zones: List[str],
+    zones: list[str],
     config: FrigateConfig,
-    target_cameras: List[str],
-) -> List[str]:
+    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
@@ -107,7 +107,7 @@ def _resolve_zones(
     if not zones:
         return zones
 
-    lookup: Dict[str, str] = {}
+    lookup: dict[str, str] = {}
     for camera_id in target_cameras:
         camera_config = config.cameras.get(camera_id)
         if camera_config is None:
@@ -120,8 +120,8 @@ def _resolve_zones(
 
 async def _execute_search_objects(
     request: Request,
-    arguments: Dict[str, Any],
-    allowed_cameras: List[str],
+    arguments: dict[str, Any],
+    allowed_cameras: list[str],
 ) -> JSONResponse:
     """
     Execute the search_objects tool.
@@ -213,8 +213,8 @@ async def _execute_search_objects(
 
 async def _execute_search_objects_semantic(
     request: Request,
-    arguments: Dict[str, Any],
-    allowed_cameras: List[str],
+    arguments: dict[str, Any],
+    allowed_cameras: list[str],
     semantic_query: str,
 ) -> JSONResponse:
     """Search objects via fused thumbnail + description embeddings.
@@ -263,8 +263,8 @@ async def _execute_search_objects_semantic(
     limit = int(arguments.get("limit", 25))
     limit = max(1, min(limit, 100))
 
-    visual_distances: Dict[str, float] = {}
-    description_distances: Dict[str, float] = {}
+    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}
@@ -305,7 +305,7 @@ async def _execute_search_objects_semantic(
 
     eligible = {e.id: e for e in Event.select().where(reduce(operator.and_, clauses))}
 
-    scored: List[tuple[str, float]] = []
+    scored: list[tuple[str, float]] = []
     for eid in eligible:
         v_score = (
             distance_to_score(visual_distances[eid], context.thumb_stats)
@@ -331,9 +331,9 @@ async def _execute_search_objects_semantic(
 
 async def _execute_find_similar_objects(
     request: Request,
-    arguments: Dict[str, Any],
-    allowed_cameras: List[str],
-) -> Dict[str, Any]:
+    arguments: dict[str, Any],
+    allowed_cameras: list[str],
+) -> dict[str, Any]:
     """Execute the find_similar_objects tool.
 
     Returns a plain dict (not JSONResponse) so the chat loop can embed it
@@ -403,8 +403,8 @@ async def _execute_find_similar_objects(
     # version (see frigate/embeddings/__init__.py). Mirror the pattern used by
     # frigate/api/event.py events_search: fetch top-k globally, then intersect
     # with the structured filters via Peewee.
-    visual_distances: Dict[str, float] = {}
-    description_distances: Dict[str, float] = {}
+    visual_distances: dict[str, float] = {}
+    description_distances: dict[str, float] = {}
 
     try:
         if similarity_mode in ("visual", "fused"):
@@ -462,7 +462,7 @@ async def _execute_find_similar_objects(
     eligible = {e.id: e for e in Event.select().where(reduce(operator.and_, clauses))}
 
     # 6. Fuse and rank.
-    scored: List[tuple[str, float]] = []
+    scored: list[tuple[str, float]] = []
     for eid in eligible:
         v_score = (
             distance_to_score(visual_distances[eid], context.thumb_stats)
@@ -503,7 +503,7 @@ async def _execute_find_similar_objects(
 async def execute_tool(
     request: Request,
     body: ToolExecuteRequest = Body(...),
-    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+    allowed_cameras: list[str] = Depends(get_allowed_cameras_for_filter),
 ) -> JSONResponse:
     """
     Execute a tool function call.
@@ -545,8 +545,8 @@ async def execute_tool(
 async def _execute_get_live_context(
     request: Request,
     camera: str,
-    allowed_cameras: List[str],
-) -> Dict[str, Any]:
+    allowed_cameras: list[str],
+) -> dict[str, Any]:
     # Reject wildcards explicitly so models retry with a real camera name
     # instead of silently fanning out across every camera.
     if camera in ("*", "all"):
@@ -593,7 +593,7 @@ async def _execute_get_live_context(
                     "stationary": obj_dict.get("stationary", False),
                 }
 
-        result: Dict[str, Any] = {
+        result: dict[str, Any] = {
             "camera": camera,
             "timestamp": frame_time,
             "detections": list(tracked_objects_dict.values()),
@@ -620,7 +620,7 @@ async def _execute_get_live_context(
 async def _get_live_frame_image_url(
     request: Request,
     camera: str,
-    allowed_cameras: List[str],
+    allowed_cameras: list[str],
 ) -> Optional[str]:
     """
     Fetch the current live frame for a camera as a base64 data URL.
@@ -659,8 +659,8 @@ async def _get_live_frame_image_url(
 
 async def _execute_set_camera_state(
     request: Request,
-    arguments: Dict[str, Any],
-) -> Dict[str, Any]:
+    arguments: dict[str, Any],
+) -> dict[str, Any]:
     role = request.headers.get("remote-role", "")
     if "admin" not in [r.strip() for r in role.split(",")]:
         return {"error": "Admin privileges required to change camera settings."}
@@ -699,10 +699,10 @@ async def _execute_set_camera_state(
 
 async def _execute_tool_internal(
     tool_name: str,
-    arguments: Dict[str, Any],
+    arguments: dict[str, Any],
     request: Request,
-    allowed_cameras: List[str],
-) -> Dict[str, Any]:
+    allowed_cameras: list[str],
+) -> dict[str, Any]:
     """
     Internal helper to execute a tool and return the result as a dict.
 
@@ -763,8 +763,8 @@ async def _execute_tool_internal(
 
 async def _execute_start_camera_watch(
     request: Request,
-    arguments: Dict[str, Any],
-) -> Dict[str, Any]:
+    arguments: dict[str, Any],
+) -> dict[str, Any]:
     camera = arguments.get("camera", "").strip()
     condition = arguments.get("condition", "").strip()
     max_duration_minutes = int(arguments.get("max_duration_minutes", 60))
@@ -814,14 +814,14 @@ async def _execute_start_camera_watch(
     }
 
 
-def _execute_stop_camera_watch() -> Dict[str, Any]:
+def _execute_stop_camera_watch() -> dict[str, Any]:
     cancelled = stop_vlm_watch_job()
     if cancelled:
         return {"success": True, "message": "Watch job cancelled."}
     return {"success": False, "message": "No active watch job to cancel."}
 
 
-def _execute_get_profile_status(request: Request) -> Dict[str, Any]:
+def _execute_get_profile_status(request: Request) -> dict[str, Any]:
     """Return profile status including active profile and activation timestamps."""
     profile_manager = getattr(request.app, "profile_manager", None)
     if profile_manager is None:
@@ -846,9 +846,9 @@ def _execute_get_profile_status(request: Request) -> Dict[str, Any]:
 
 
 def _execute_get_recap(
-    arguments: Dict[str, Any],
-    allowed_cameras: List[str],
-) -> Dict[str, Any]:
+    arguments: dict[str, Any],
+    allowed_cameras: list[str],
+) -> dict[str, Any]:
     """Fetch review segments with GenAI metadata for a time period."""
     from functools import reduce
 
@@ -909,7 +909,7 @@ def _execute_get_recap(
             .iterator()
         )
 
-        events: List[Dict[str, Any]] = []
+        events: list[dict[str, Any]] = []
 
         for row in rows:
             data = row.get("data") or {}
@@ -920,7 +920,7 @@ def _execute_get_recap(
                     data = {}
 
             camera = row["camera"]
-            event: Dict[str, Any] = {
+            event: dict[str, Any] = {
                 "camera": camera.replace("_", " ").title(),
                 "severity": row.get("severity", "detection"),
             }
@@ -984,10 +984,10 @@ def _execute_get_recap(
 
 
 async def _execute_pending_tools(
-    pending_tool_calls: List[Dict[str, Any]],
+    pending_tool_calls: list[dict[str, Any]],
     request: Request,
-    allowed_cameras: List[str],
-) -> tuple[List[ToolCall], List[Dict[str, Any]], List[Dict[str, Any]]]:
+    allowed_cameras: list[str],
+) -> tuple[list[ToolCall], list[dict[str, Any]], list[dict[str, Any]]]:
     """
     Execute a list of tool calls.
 
@@ -996,9 +996,9 @@ async def _execute_pending_tools(
          tool result dicts for conversation,
          extra messages to inject after tool results — e.g. user messages with images)
     """
-    tool_calls_out: List[ToolCall] = []
-    tool_results: List[Dict[str, Any]] = []
-    extra_messages: List[Dict[str, Any]] = []
+    tool_calls_out: list[ToolCall] = []
+    tool_results: list[dict[str, Any]] = []
+    extra_messages: list[dict[str, Any]] = []
     for tool_call in pending_tool_calls:
         tool_name = tool_call["name"]
         tool_args = tool_call.get("arguments") or {}
@@ -1106,7 +1106,7 @@ async def _execute_pending_tools(
 async def chat_completion(
     request: Request,
     body: ChatCompletionRequest = Body(...),
-    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
+    allowed_cameras: list[str] = Depends(get_allowed_cameras_for_filter),
 ):
     """
     Chat completion endpoint with tool calling support.
@@ -1138,19 +1138,23 @@ async def chat_completion(
     )
     conversation = []
 
-    system_prompt = build_chat_system_prompt(
-        config=config,
-        allowed_cameras=allowed_cameras,
-        semantic_search_enabled=semantic_search_enabled,
-        attribute_classifications=attribute_classifications,
-    )
-
-    conversation.append(
-        {
-            "role": "system",
-            "content": system_prompt,
-        }
-    )
+    # Build the system message only when the client hasn't already pinned one.
+    # The first turn has no system message; we generate it (with the current
+    # timestamp) and return the whole chain so the client persists it. Later
+    # turns send it back verbatim, freezing the timestamp so the prompt prefix
+    # stays byte-identical and the model server's prompt cache keeps hitting.
+    if not body.messages or body.messages[0].role != "system":
+        conversation.append(
+            {
+                "role": "system",
+                "content": build_chat_system_prompt(
+                    config=config,
+                    allowed_cameras=allowed_cameras,
+                    semantic_search_enabled=semantic_search_enabled,
+                    attribute_classifications=attribute_classifications,
+                ),
+            }
+        )
 
     for msg in body.messages:
         msg_dict = {
@@ -1161,11 +1165,13 @@ async def chat_completion(
             msg_dict["tool_call_id"] = msg.tool_call_id
         if msg.name:
             msg_dict["name"] = msg.name
+        if msg.tool_calls is not None:
+            msg_dict["tool_calls"] = msg.tool_calls
 
         conversation.append(msg_dict)
 
     tool_iterations = 0
-    tool_calls: List[ToolCall] = []
+    tool_calls: list[ToolCall] = []
     max_iterations = body.max_tool_iterations
 
     logger.debug(
@@ -1175,11 +1181,20 @@ async def chat_completion(
 
     # True LLM streaming when client supports it and stream requested
     if body.stream and hasattr(genai_client, "chat_with_tools_stream"):
-        stream_tool_calls: List[ToolCall] = []
         stream_iterations = 0
 
         async def stream_body_llm():
-            nonlocal conversation, stream_tool_calls, stream_iterations
+            nonlocal conversation, stream_iterations
+
+            def _emit_chain(extra: Optional[list[dict[str, Any]]] = None):
+                # Return the full conversation (including the system message) so
+                # the client persists and replays it verbatim next turn.
+                chain = conversation + (extra or [])
+                return (
+                    json.dumps({"type": "messages", "messages": chain}).encode("utf-8")
+                    + b"\n"
+                )
+
             while stream_iterations < max_iterations:
                 if await request.is_disconnected():
                     logger.debug("Client disconnected, stopping chat stream")
@@ -1244,31 +1259,33 @@ async def chat_completion(
                                 )
                                 return
                             (
-                                executed_calls,
+                                _executed_calls,
                                 tool_results,
                                 extra_msgs,
                             ) = await _execute_pending_tools(
                                 pending, request, allowed_cameras
                             )
-                            stream_tool_calls.extend(executed_calls)
                             conversation.extend(tool_results)
                             conversation.extend(extra_msgs)
-                            yield (
-                                json.dumps(
-                                    {
-                                        "type": "tool_calls",
-                                        "tool_calls": [
-                                            tc.model_dump() for tc in stream_tool_calls
-                                        ],
-                                    }
-                                ).encode("utf-8")
-                                + b"\n"
-                            )
+                            # Emit the running chain so the client can render tool
+                            # calls live and replay them verbatim next turn.
+                            yield _emit_chain()
                             break
                         else:
+                            # Streaming never appends the final assistant message
+                            # to the conversation, so add it to the chain.
+                            yield _emit_chain(
+                                extra=[
+                                    {
+                                        "role": "assistant",
+                                        "content": msg.get("content"),
+                                    }
+                                ]
+                            )
                             yield (json.dumps({"type": "done"}).encode("utf-8") + b"\n")
                             return
             else:
+                yield _emit_chain()
                 yield json.dumps({"type": "done"}).encode("utf-8") + b"\n"
 
         return StreamingResponse(
@@ -1315,19 +1332,15 @@ async def chat_completion(
                 if body.stream:
                     final_reasoning = response.get("reasoning")
 
+                    chain = list(conversation)
+
                     async def stream_body() -> Any:
-                        if tool_calls:
-                            yield (
-                                json.dumps(
-                                    {
-                                        "type": "tool_calls",
-                                        "tool_calls": [
-                                            tc.model_dump() for tc in tool_calls
-                                        ],
-                                    }
-                                ).encode("utf-8")
-                                + b"\n"
+                        yield (
+                            json.dumps({"type": "messages", "messages": chain}).encode(
+                                "utf-8"
                             )
+                            + b"\n"
+                        )
                         # Emit the full reasoning trace up front when the
                         # underlying client did not stream it
                         if final_reasoning:
@@ -1363,6 +1376,7 @@ async def chat_completion(
                         finish_reason=response.get("finish_reason", "stop"),
                         tool_iterations=tool_iterations,
                         tool_calls=tool_calls,
+                        messages=list(conversation),
                     ).model_dump(),
                 )
 
@@ -1395,6 +1409,7 @@ async def chat_completion(
                 finish_reason="length",
                 tool_iterations=tool_iterations,
                 tool_calls=tool_calls,
+                messages=list(conversation),
             ).model_dump(),
         )
 

frigate/api/defs/request/chat_body.py

@@ -1,6 +1,6 @@
 """Chat API request models."""
 
-from typing import Optional
+from typing import Any, Optional
 
 from pydantic import BaseModel, Field
 
@@ -11,13 +11,29 @@ class ChatMessage(BaseModel):
     role: str = Field(
         description="Message role: 'user', 'assistant', 'system', or 'tool'"
     )
-    content: str = Field(description="Message content")
+    content: Optional[Any] = Field(
+        default=None,
+        description=(
+            "Message content. Usually a string, but may be a multimodal content "
+            "list (e.g. text + image_url) or null for assistant turns that only "
+            "request tool calls."
+        ),
+    )
     tool_call_id: Optional[str] = Field(
         default=None, description="For tool messages, the ID of the tool call"
     )
     name: Optional[str] = Field(
         default=None, description="For tool messages, the tool name"
     )
+    tool_calls: Optional[list[dict[str, Any]]] = Field(
+        default=None,
+        description=(
+            "For assistant messages replayed from prior turns, the OpenAI-format "
+            "tool calls the model previously requested. Replaying these verbatim "
+            "keeps the conversation prefix byte-for-byte identical so the model "
+            "server's prompt cache hits on follow-up turns."
+        ),
+    )
 
 
 class ChatCompletionRequest(BaseModel):

frigate/api/defs/response/chat_response.py

@@ -56,3 +56,12 @@ class ChatCompletionResponse(BaseModel):
         default_factory=list,
         description="List of tool calls that were executed during this completion",
     )
+    messages: list[dict[str, Any]] = Field(
+        default_factory=list,
+        description=(
+            "The full conversation chain, including the system message. Persist "
+            "and replay this verbatim on the next request so the prompt prefix "
+            "stays byte-identical and the model server's prompt cache keeps "
+            "hitting."
+        ),
+    )

frigate/config/camera/updater.py

@@ -73,7 +73,12 @@ class CameraConfigUpdateSubscriber:
 
         base_topic = "config/cameras"
 
-        if len(self.camera_configs) == 1:
+        # global subscribers must hear every camera; only narrow per-camera workers
+        is_global_subscriber = (
+            CameraConfigUpdateEnum.add in self.topics
+            or CameraConfigUpdateEnum.remove in self.topics
+        )
+        if not is_global_subscriber and len(self.camera_configs) == 1:
             base_topic += f"/{list(self.camera_configs.keys())[0]}"
 
         self.subscriber = ConfigSubscriber(

frigate/test/http_api/test_http_keyframe_analysis.py

@@ -0,0 +1,58 @@
+from unittest.mock import AsyncMock, patch
+
+from frigate.models import Event, Recordings, ReviewSegment
+from frigate.test.http_api.base_http_test import AuthTestClient, BaseTestHttp
+
+
+class TestHttpKeyframeAnalysis(BaseTestHttp):
+    def setUp(self):
+        super().setUp([Event, Recordings, ReviewSegment])
+
+    def test_invalid_camera_returns_404(self):
+        app = super().create_app()
+        with AuthTestClient(app) as client:
+            response = client.get("/keyframe_analysis?camera=does_not_exist")
+            assert response.status_code == 404
+
+    def test_record_disabled_returns_neutral(self):
+        # default minimal_config has recording disabled
+        app = super().create_app()
+        with AuthTestClient(app) as client:
+            response = client.get("/keyframe_analysis?camera=front_door")
+            assert response.status_code == 200
+            assert response.json()["severity"] == "record_disabled"
+
+    def test_probes_record_input_and_returns_severity(self):
+        self.minimal_config["cameras"]["front_door"]["ffmpeg"]["inputs"] = [
+            {
+                "path": "rtsp://10.0.0.1:554/record",
+                "roles": ["detect", "record"],
+            }
+        ]
+        self.minimal_config["cameras"]["front_door"]["record"] = {"enabled": True}
+        app = super().create_app()
+
+        canned = {
+            "severity": "ok",
+            "keyframe_count": 5,
+            "max_gap": 1.0,
+            "mean_gap": 1.0,
+            "min_gap": 1.0,
+            "segment_time": 10,
+            "duration_observed": 4.0,
+            "thresholds": {"warning": 4.0, "error": 10},
+        }
+
+        with patch(
+            "frigate.api.camera.analyze_record_keyframes",
+            AsyncMock(return_value=canned),
+        ) as mock_probe:
+            with AuthTestClient(app) as client:
+                response = client.get("/keyframe_analysis?camera=front_door")
+
+        assert response.status_code == 200
+        assert response.json()["severity"] == "ok"
+        # index matches the input carrying the record role ("Stream 1")
+        assert response.json()["stream_index"] == 0
+        # the record-role input path was probed
+        assert mock_probe.await_args.args[1] == "rtsp://10.0.0.1:554/record"

frigate/test/test_keyframe_analysis.py

@@ -0,0 +1,111 @@
+"""Tests for keyframe-spacing analysis used to detect smart/+ codecs."""
+
+import asyncio
+import unittest
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from frigate.util.services import (
+    analyze_record_keyframes,
+    classify_keyframe_gaps,
+    parse_keyframe_packets,
+)
+
+
+class TestClassifyKeyframeGaps(unittest.TestCase):
+    def test_ok_when_gaps_small(self):
+        # keyframes every ~1s
+        pts = [0.0, 1.0, 2.0, 3.0, 4.0]
+        result = classify_keyframe_gaps(pts, segment_time=10)
+        self.assertEqual(result["severity"], "ok")
+        self.assertEqual(result["max_gap"], 1.0)
+        self.assertEqual(result["keyframe_count"], 5)
+        self.assertEqual(result["thresholds"], {"warning": 4.0, "error": 10})
+
+    def test_warning_when_gap_exceeds_four_seconds(self):
+        pts = [0.0, 1.0, 6.5]  # 5.5s gap
+        result = classify_keyframe_gaps(pts, segment_time=10)
+        self.assertEqual(result["severity"], "warning")
+        self.assertEqual(result["max_gap"], 5.5)
+
+    def test_error_when_gap_exceeds_segment_time(self):
+        pts = [0.0, 12.0]  # 12s gap > 10s segment
+        result = classify_keyframe_gaps(pts, segment_time=10)
+        self.assertEqual(result["severity"], "error")
+
+    def test_error_threshold_tracks_segment_time(self):
+        pts = [0.0, 6.0]  # 6s gap, segment_time=5 -> error
+        result = classify_keyframe_gaps(pts, segment_time=5)
+        self.assertEqual(result["severity"], "error")
+
+    def test_unknown_with_single_keyframe(self):
+        result = classify_keyframe_gaps([1.0], segment_time=10)
+        self.assertEqual(result["severity"], "unknown")
+        self.assertIsNone(result["max_gap"])
+        self.assertEqual(result["keyframe_count"], 1)
+
+    def test_unknown_with_no_keyframes(self):
+        result = classify_keyframe_gaps([], segment_time=10)
+        self.assertEqual(result["severity"], "unknown")
+        self.assertEqual(result["keyframe_count"], 0)
+
+
+class TestParseKeyframePackets(unittest.TestCase):
+    def test_extracts_keyframe_pts_and_max(self):
+        output = "0.000000,K__\n0.033333,___\n1.000000,K__\n1.500000,___\n"
+        keyframe_pts, max_pts = parse_keyframe_packets(output)
+        self.assertEqual(keyframe_pts, [0.0, 1.0])
+        self.assertEqual(max_pts, 1.5)
+
+    def test_skips_unparseable_and_empty_lines(self):
+        output = "N/A,K__\n\n2.0,K__\nbad line\n"
+        keyframe_pts, max_pts = parse_keyframe_packets(output)
+        self.assertEqual(keyframe_pts, [2.0])
+        self.assertEqual(max_pts, 2.0)
+
+    def test_empty_output(self):
+        keyframe_pts, max_pts = parse_keyframe_packets("")
+        self.assertEqual(keyframe_pts, [])
+        self.assertIsNone(max_pts)
+
+
+class TestAnalyzeRecordKeyframes(unittest.IsolatedAsyncioTestCase):
+    async def test_merges_duration_and_classification(self):
+        csv = b"0.0,K__\n1.0,___\n6.0,K__\n7.0,___\n"
+        proc = MagicMock()
+        proc.communicate = AsyncMock(return_value=(csv, b""))
+        ffmpeg = MagicMock()
+        ffmpeg.ffprobe_path = "/usr/bin/ffprobe"
+
+        with patch(
+            "frigate.util.services.asyncio.create_subprocess_exec",
+            AsyncMock(return_value=proc),
+        ):
+            result = await analyze_record_keyframes(
+                ffmpeg, "rtsp://cam/stream", segment_time=10
+            )
+
+        self.assertEqual(result["severity"], "warning")  # 6s gap > 4s
+        self.assertEqual(result["max_gap"], 6.0)
+        self.assertEqual(result["duration_observed"], 7.0)
+
+    async def test_timeout_returns_unknown(self):
+        proc = MagicMock()
+        proc.communicate = AsyncMock(side_effect=asyncio.TimeoutError())
+        proc.kill = MagicMock()
+        ffmpeg = MagicMock()
+        ffmpeg.ffprobe_path = "/usr/bin/ffprobe"
+
+        with patch(
+            "frigate.util.services.asyncio.create_subprocess_exec",
+            AsyncMock(return_value=proc),
+        ):
+            result = await analyze_record_keyframes(
+                ffmpeg, "rtsp://cam/stream", segment_time=10
+            )
+
+        self.assertEqual(result["severity"], "unknown")
+        proc.kill.assert_called_once()
+
+
+if __name__ == "__main__":
+    unittest.main()

frigate/util/builtin.py

@@ -14,13 +14,16 @@ import urllib.parse
 from collections.abc import Mapping
 from multiprocessing.managers import ValueProxy
 from pathlib import Path
-from typing import Any, Dict, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
 
 import numpy as np
 from ruamel.yaml import YAML
 
 from frigate.const import REGEX_HTTP_CAMERA_USER_PASS, REGEX_RTSP_CAMERA_USER_PASS
 
+if TYPE_CHECKING:
+    from frigate.config import CameraConfig
+
 logger = logging.getLogger(__name__)
 
 
@@ -132,6 +135,24 @@ def get_ffmpeg_arg_list(arg: Any) -> list:
     return arg if isinstance(arg, list) else shlex.split(arg)
 
 
+# all built-in record presets use this segment_time
+DEFAULT_RECORD_SEGMENT_TIME = 10
+
+
+def get_record_segment_time(config: "CameraConfig") -> int:
+    """Extract -segment_time from the camera's record output args."""
+    record_args = get_ffmpeg_arg_list(config.ffmpeg.output_args.record)
+
+    if record_args and record_args[0].startswith("preset"):
+        return DEFAULT_RECORD_SEGMENT_TIME
+
+    try:
+        idx = record_args.index("-segment_time")
+        return int(record_args[idx + 1])
+    except (ValueError, IndexError):
+        return DEFAULT_RECORD_SEGMENT_TIME
+
+
 def load_labels(
     path: Optional[str], encoding="utf-8", prefill=91, indexed: bool | None = None
 ):

frigate/util/services.py

@@ -879,6 +879,131 @@ def ffprobe_stream(ffmpeg, path: str, detailed: bool = False) -> sp.CompletedPro
     return result
 
 
+KEYFRAME_PROBE_WINDOW_SECONDS = 20
+KEYFRAME_GAP_WARNING_SECONDS = 4.0
+
+
+def parse_keyframe_packets(output: str) -> Tuple[List[float], Optional[float]]:
+    """Parse ffprobe CSV `pts_time,flags` output.
+
+    Returns the presentation timestamps of keyframes (flags containing "K")
+    and the maximum timestamp observed across all packets.
+    """
+    keyframe_pts: List[float] = []
+    max_pts: Optional[float] = None
+
+    for line in output.splitlines():
+        parts = line.split(",")
+        if len(parts) < 2:
+            continue
+        try:
+            pts = float(parts[0])
+        except ValueError:
+            continue
+        if max_pts is None or pts > max_pts:
+            max_pts = pts
+        if "K" in parts[1]:
+            keyframe_pts.append(pts)
+
+    return keyframe_pts, max_pts
+
+
+def classify_keyframe_gaps(
+    keyframe_pts: List[float], segment_time: int
+) -> dict[str, Any]:
+    """Classify keyframe spacing for recording suitability.
+
+    A camera using a smart/+ codec or a long/variable GOP produces large or
+    irregular gaps between keyframes, which breaks time-based recording
+    segmentation. Severity:
+      - "unknown" when fewer than two keyframes were observed
+      - "error" when the longest gap exceeds the record segment length
+      - "warning" when the longest gap exceeds the warning threshold
+      - "ok" otherwise
+    """
+    thresholds = {
+        "warning": KEYFRAME_GAP_WARNING_SECONDS,
+        "error": segment_time,
+    }
+
+    if len(keyframe_pts) < 2:
+        return {
+            "keyframe_count": len(keyframe_pts),
+            "max_gap": None,
+            "mean_gap": None,
+            "min_gap": None,
+            "segment_time": segment_time,
+            "severity": "unknown",
+            "thresholds": thresholds,
+        }
+
+    gaps = [b - a for a, b in zip(keyframe_pts, keyframe_pts[1:])]
+    max_gap = max(gaps)
+
+    if max_gap > segment_time:
+        severity = "error"
+    elif max_gap > KEYFRAME_GAP_WARNING_SECONDS:
+        severity = "warning"
+    else:
+        severity = "ok"
+
+    return {
+        "keyframe_count": len(keyframe_pts),
+        "max_gap": round(max_gap, 2),
+        "mean_gap": round(sum(gaps) / len(gaps), 2),
+        "min_gap": round(min(gaps), 2),
+        "segment_time": segment_time,
+        "severity": severity,
+        "thresholds": thresholds,
+    }
+
+
+async def analyze_record_keyframes(
+    ffmpeg, url: str, segment_time: int, window: int = KEYFRAME_PROBE_WINDOW_SECONDS
+) -> dict[str, Any]:
+    """Probe a stream for ~`window` seconds and classify its keyframe spacing.
+
+    Reads video packet flags via ffprobe to find keyframes, then measures the
+    gaps between them. On timeout or failure returns an "unknown" result rather
+    than a false all-clear.
+    """
+    clean_url = escape_special_characters(url)
+    cmd = [
+        ffmpeg.ffprobe_path,
+        "-v",
+        "error",
+        "-select_streams",
+        "v:0",
+        "-read_intervals",
+        f"%+{window}",
+        "-show_entries",
+        "packet=pts_time,flags",
+        "-of",
+        "csv=p=0",
+        clean_url,
+    ]
+
+    try:
+        proc = await asyncio.create_subprocess_exec(
+            *cmd,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+        stdout, _ = await asyncio.wait_for(proc.communicate(), timeout=window + 15)
+    except asyncio.TimeoutError:
+        logger.warning("Keyframe probe timed out for record stream")
+        proc.kill()
+        return classify_keyframe_gaps([], segment_time)
+    except OSError as err:
+        logger.error("Keyframe probe failed: %s", err)
+        return classify_keyframe_gaps([], segment_time)
+
+    keyframe_pts, max_pts = parse_keyframe_packets(stdout.decode("utf-8", "replace"))
+    result = classify_keyframe_gaps(keyframe_pts, segment_time)
+    result["duration_observed"] = round(max_pts, 2) if max_pts is not None else None
+    return result
+
+
 def vainfo_hwaccel(device_name: Optional[str] = None) -> sp.CompletedProcess:
     """Run vainfo."""
     if not device_name:

frigate/video/ffmpeg.py

@@ -24,7 +24,7 @@ from frigate.config.camera.updater import (
 )
 from frigate.const import PROCESS_PRIORITY_HIGH
 from frigate.log import LogPipe
-from frigate.util.builtin import EventsPerSecond, get_ffmpeg_arg_list
+from frigate.util.builtin import EventsPerSecond, get_record_segment_time
 from frigate.util.ffmpeg import start_or_restart_ffmpeg, stop_ffmpeg
 from frigate.util.image import (
     FrameManager,
@@ -34,23 +34,6 @@ from frigate.util.process import FrigateProcess
 
 logger = logging.getLogger(__name__)
 
-# all built-in record presets use this segment_time
-DEFAULT_RECORD_SEGMENT_TIME = 10
-
-
-def _get_record_segment_time(config: CameraConfig) -> int:
-    """Extract -segment_time from the camera's record output args."""
-    record_args = get_ffmpeg_arg_list(config.ffmpeg.output_args.record)
-
-    if record_args and record_args[0].startswith("preset"):
-        return DEFAULT_RECORD_SEGMENT_TIME
-
-    try:
-        idx = record_args.index("-segment_time")
-        return int(record_args[idx + 1])
-    except (ValueError, IndexError):
-        return DEFAULT_RECORD_SEGMENT_TIME
-
 
 def capture_frames(
     ffmpeg_process: sp.Popen[Any],
@@ -185,7 +168,7 @@ class CameraWatchdog(threading.Thread):
         # `valid` segments are published with the segment's start time, so the
         # gap between consecutive publishes can reach 2 * segment_time. Pad the
         # staleness threshold so it's never tighter than that worst case.
-        segment_time = _get_record_segment_time(self.config)
+        segment_time = get_record_segment_time(self.config)
         self.record_stale_threshold = max(120, 2 * segment_time + 30)
 
         # Stall tracking (based on last processed frame)

web/.gitignore

@@ -12,6 +12,10 @@ dist
 dist-ssr
 *.local
 
+# Playwright
+playwright-report
+test-results
+
 # Editor directories and files
 .vscode/*
 !.vscode/extensions.json

web/e2e/specs/chat.spec.ts

@@ -92,6 +92,15 @@ test.describe("Chat — streaming @medium", () => {
     await installChatStreamOverride(frigateApp, [
       { type: "content", delta: "Hel" },
       { type: "content", delta: "lo" },
+      {
+        type: "messages",
+        messages: [
+          { role: "system", content: "sys" },
+          { role: "user", content: "hello chat" },
+          { role: "assistant", content: "Hello" },
+        ],
+      },
+      { type: "done" },
     ]);
     await frigateApp.goto("/chat");
     const input = frigateApp.page.getByPlaceholder(/ask/i);
@@ -137,6 +146,15 @@ test.describe("Chat — streaming @medium", () => {
         { type: "content", delta: "Hel" },
         { type: "content", delta: "lo, " },
         { type: "content", delta: "world!" },
+        {
+          type: "messages",
+          messages: [
+            { role: "system", content: "sys" },
+            { role: "user", content: "greet me" },
+            { role: "assistant", content: "Hello, world!" },
+          ],
+        },
+        { type: "done" },
       ],
       { chunkDelayMs: 50 },
     );
@@ -151,19 +169,39 @@ test.describe("Chat — streaming @medium", () => {
     });
   });
 
-  test("tool_calls chunks render a ToolCallsGroup", async ({ frigateApp }) => {
-    await installChatStreamOverride(frigateApp, [
+  test("tool calls in the chain render a ToolCallsGroup", async ({
+    frigateApp,
+  }) => {
+    const toolTurn = [
+      { role: "system", content: "sys" },
+      { role: "user", content: "find people" },
       {
-        type: "tool_calls",
+        role: "assistant",
+        content: null,
         tool_calls: [
           {
             id: "call_1",
-            name: "search_objects",
-            arguments: { label: "person" },
+            type: "function",
+            function: {
+              name: "search_objects",
+              arguments: '{"label":"person"}',
+            },
           },
         ],
       },
+      { role: "tool", tool_call_id: "call_1", content: "[]" },
+    ];
+    await installChatStreamOverride(frigateApp, [
+      { type: "messages", messages: toolTurn },
       { type: "content", delta: "Searching for people." },
+      {
+        type: "messages",
+        messages: [
+          ...toolTurn,
+          { role: "assistant", content: "Searching for people." },
+        ],
+      },
+      { type: "done" },
     ]);
     await frigateApp.goto("/chat");
     const input = frigateApp.page.getByPlaceholder(/ask/i);
@@ -253,6 +291,15 @@ test.describe("Chat — attachment chip @medium", () => {
     // We use the stream override so the first message completes quickly.
     await installChatStreamOverride(frigateApp, [
       { type: "content", delta: "Done." },
+      {
+        type: "messages",
+        messages: [
+          { role: "system", content: "sys" },
+          { role: "user", content: "hello" },
+          { role: "assistant", content: "Done." },
+        ],
+      },
+      { type: "done" },
     ]);
     await frigateApp.goto("/chat");
 

web/public/locales/en/views/system.json

@@ -174,6 +174,21 @@
       "error": "Error: {{error}}",
       "tips": {
         "title": "Camera Probe Info"
+      },
+      "keyframes": {
+        "title": "Keyframe analysis",
+        "analyzing": "Analyzing keyframes... {{seconds}} seconds remaining",
+        "stillAnalyzing": "Still analyzing keyframes...",
+        "recordStream": "Record stream:",
+        "keyframeCount": "Keyframes observed:",
+        "observedDuration": "Observed duration:",
+        "gap": "Keyframe gap (min / avg / max):",
+        "segmentLength": "Recording segment length:",
+        "ok": "Keyframes every ~{{seconds}}s, good for recording and playback.",
+        "warning": "Sparse or variable keyframes (longest gap ~{{seconds}}s), likely a smart codec (H.264+/H.265+), this is not recommended.",
+        "error": "Keyframe gap (~{{seconds}}s) exceeds the recording segment length ({{segmentTime}}s). Some segments may have no keyframe, which breaks playback. Disable the smart/+ codec on the camera or shorten its keyframe interval.",
+        "unknown": "Couldn't determine keyframe spacing.",
+        "recordDisabled": "Recording is disabled for this camera."
       }
     },
     "framesAndDetections": "Frames / Detections",

web/src/components/overlay/CameraInfoDialog.tsx

@@ -7,7 +7,8 @@ import {
   DialogTitle,
 } from "../ui/dialog";
 import ActivityIndicator from "../indicators/activity-indicator";
-import { Ffprobe } from "@/types/stats";
+import KeyframeAnalysisSection from "./KeyframeAnalysisSection";
+import { Ffprobe, KeyframeAnalysis } from "@/types/stats";
 import { Button } from "../ui/button";
 import copy from "copy-to-clipboard";
 import { CameraConfig } from "@/types/frigateConfig";
@@ -30,6 +31,7 @@ export default function CameraInfoDialog({
 }: CameraInfoDialogProps) {
   const { t } = useTranslation(["views/system"]);
   const [ffprobeInfo, setFfprobeInfo] = useState<Ffprobe[]>();
+  const [keyframeInfo, setKeyframeInfo] = useState<KeyframeAnalysis>();
 
   useEffect(() => {
     axios
@@ -67,7 +69,12 @@ export default function CameraInfoDialog({
   }, []);
 
   const onCopyFfprobe = async () => {
-    copy(JSON.stringify(ffprobeInfo));
+    copy(
+      JSON.stringify({
+        ffprobe: ffprobeInfo,
+        keyframe_analysis: keyframeInfo,
+      }),
+    );
     toast.success(t("cameras.toast.success.copyToClipboard"));
   };
 
@@ -96,7 +103,7 @@ export default function CameraInfoDialog({
             <Trans ns="views/system">cameras.info.streamDataFromFFPROBE</Trans>
           </DialogDescription>
 
-          <div className="mb-2 p-4">
+          <div className="mb-2 p-4 text-sm">
             {ffprobeInfo ? (
               <div>
                 {ffprobeInfo.map((stream, idx) => (
@@ -184,6 +191,10 @@ export default function CameraInfoDialog({
                     )}
                   </div>
                 ))}
+                <KeyframeAnalysisSection
+                  cameraName={camera.name}
+                  onResult={setKeyframeInfo}
+                />
               </div>
             ) : (
               <div className="flex flex-col items-center">

web/src/components/overlay/KeyframeAnalysisSection.tsx

@@ -0,0 +1,193 @@
+import { useEffect, useMemo, useState } from "react";
+import { useTranslation } from "react-i18next";
+import axios from "axios";
+import { FaCircleCheck, FaTriangleExclamation } from "react-icons/fa6";
+import { LuX } from "react-icons/lu";
+import ActivityIndicator from "../indicators/activity-indicator";
+import { KeyframeAnalysis } from "@/types/stats";
+
+const PROBE_WINDOW_SECONDS = 20;
+
+type KeyframeAnalysisSectionProps = {
+  cameraName: string;
+  onResult?: (analysis: KeyframeAnalysis) => void;
+};
+
+export default function KeyframeAnalysisSection({
+  cameraName,
+  onResult,
+}: KeyframeAnalysisSectionProps) {
+  const { t } = useTranslation(["views/system"]);
+  const [analysis, setAnalysis] = useState<KeyframeAnalysis>();
+  const [failed, setFailed] = useState(false);
+  const [secondsRemaining, setSecondsRemaining] =
+    useState(PROBE_WINDOW_SECONDS);
+
+  // fire the probe once on mount
+  useEffect(() => {
+    let active = true;
+    axios
+      .get("keyframe_analysis", { params: { camera: cameraName } })
+      .then((res) => {
+        if (active) {
+          setAnalysis(res.data);
+          onResult?.(res.data);
+        }
+      })
+      .catch(() => {
+        if (active) {
+          setFailed(true);
+        }
+      });
+    return () => {
+      active = false;
+    };
+    // re-probing only depends on the camera; onResult is a stable setter
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [cameraName]);
+
+  // countdown while waiting for the probe to return
+  useEffect(() => {
+    if (analysis || failed) {
+      return;
+    }
+    const interval = setInterval(() => {
+      setSecondsRemaining((s) => (s > 0 ? s - 1 : 0));
+    }, 1000);
+    return () => clearInterval(interval);
+  }, [analysis, failed]);
+
+  const content = useMemo(() => {
+    if (failed) {
+      return <Row icon="unknown">{t("cameras.info.keyframes.unknown")}</Row>;
+    }
+
+    if (!analysis) {
+      return (
+        <div className="flex items-center gap-2 text-muted-foreground">
+          <ActivityIndicator className="size-4" />
+          <span>
+            {secondsRemaining > 0
+              ? t("cameras.info.keyframes.analyzing", {
+                  seconds: secondsRemaining,
+                })
+              : t("cameras.info.keyframes.stillAnalyzing")}
+          </span>
+        </div>
+      );
+    }
+
+    let summary;
+    switch (analysis.severity) {
+      case "ok":
+        summary = (
+          <Row icon="ok">
+            {t("cameras.info.keyframes.ok", { seconds: analysis.mean_gap })}
+          </Row>
+        );
+        break;
+      case "warning":
+        summary = (
+          <Row icon="warning">
+            {t("cameras.info.keyframes.warning", { seconds: analysis.max_gap })}
+          </Row>
+        );
+        break;
+      case "error":
+        summary = (
+          <Row icon="error">
+            {t("cameras.info.keyframes.error", {
+              seconds: analysis.max_gap,
+              segmentTime: analysis.segment_time,
+            })}
+          </Row>
+        );
+        break;
+      case "record_disabled":
+        summary = (
+          <Row icon="unknown">{t("cameras.info.keyframes.recordDisabled")}</Row>
+        );
+        break;
+      default:
+        summary = (
+          <Row icon="unknown">{t("cameras.info.keyframes.unknown")}</Row>
+        );
+    }
+
+    // gap statistics are only meaningful once at least two keyframes were seen
+    const hasStats = analysis.max_gap != null;
+    const hasDetails = hasStats || analysis.stream_index != null;
+
+    return (
+      <div className="text-muted-foreground">
+        {analysis.stream_index != null && (
+          <div>
+            {t("cameras.info.keyframes.recordStream")}{" "}
+            <span className="text-primary">
+              {t("cameras.info.stream", { idx: analysis.stream_index + 1 })}
+            </span>
+          </div>
+        )}
+        {hasStats && (
+          <div>
+            <div>
+              {t("cameras.info.keyframes.keyframeCount")}{" "}
+              <span className="text-primary">{analysis.keyframe_count}</span>
+            </div>
+            <div>
+              {t("cameras.info.keyframes.observedDuration")}{" "}
+              <span className="text-primary">
+                {analysis.duration_observed}s
+              </span>
+            </div>
+            <div>
+              {t("cameras.info.keyframes.gap")}{" "}
+              <span className="text-primary">
+                {analysis.min_gap}s / {analysis.mean_gap}s / {analysis.max_gap}s
+              </span>
+            </div>
+            <div>
+              {t("cameras.info.keyframes.segmentLength")}{" "}
+              <span className="text-primary">{analysis.segment_time}s</span>
+            </div>
+          </div>
+        )}
+        <div className={hasDetails ? "mt-3" : undefined}>{summary}</div>
+      </div>
+    );
+  }, [analysis, failed, secondsRemaining, t]);
+
+  return (
+    <div className="mb-5">
+      <div className="mb-1 rounded-md bg-secondary p-2 text-lg text-primary">
+        {t("cameras.info.keyframes.title")}
+      </div>
+      <div className="ml-2">{content}</div>
+    </div>
+  );
+}
+
+type RowProps = {
+  icon: "ok" | "warning" | "error" | "unknown";
+  children: React.ReactNode;
+};
+
+function Row({ icon, children }: RowProps) {
+  return (
+    <div className="flex items-start gap-2">
+      {icon === "ok" && (
+        <FaCircleCheck className="mt-0.5 size-4 flex-shrink-0 text-success" />
+      )}
+      {icon === "warning" && (
+        <FaTriangleExclamation className="mt-0.5 size-4 flex-shrink-0 text-yellow-500" />
+      )}
+      {icon === "error" && (
+        <LuX className="mt-0.5 size-4 flex-shrink-0 text-danger" />
+      )}
+      {icon === "unknown" && (
+        <FaTriangleExclamation className="mt-0.5 size-4 flex-shrink-0 text-muted-foreground" />
+      )}
+      <span className="text-primary">{children}</span>
+    </div>
+  );
+}

web/src/pages/Chat.tsx

@@ -13,6 +13,7 @@ import { ChatComposer } from "@/components/chat/ChatComposer";
 import ChatSettings from "@/components/chat/ChatSettings";
 import type {
   ChatMessage,
+  ChatStats,
   GenAIModelsResponse,
   ShowStatsMode,
 } from "@/types/chat";
@@ -22,12 +23,28 @@ import {
   getFindSimilarObjectsFromToolCalls,
   prependAttachment,
   streamChatCompletion,
+  toolCallsForMessage,
+  toolResponsesById,
 } from "@/utils/chatUtil";
 
+type StreamingTurn = {
+  content: string;
+  reasoning: string;
+  chain: ChatMessage[];
+  stats?: ChatStats;
+};
+
+const hasText = (content: unknown): content is string =>
+  typeof content === "string" && content.trim().length > 0;
+
+const toWire = (messages: ChatMessage[]): ChatMessage[] =>
+  messages.map(({ reasoning: _r, stats: _s, ...rest }) => rest);
+
 export default function ChatPage() {
   const { t } = useTranslation(["views/chat"]);
   const [input, setInput] = useState("");
   const [messages, setMessages] = useState<ChatMessage[]>([]);
+  const [streaming, setStreaming] = useState<StreamingTurn | null>(null);
   const [isLoading, setIsLoading] = useState(false);
   const [error, setError] = useState<string | null>(null);
   const [attachedEventId, setAttachedEventId] = useState<string | null>(null);
@@ -72,28 +89,19 @@ export default function ChatPage() {
     if (isNearBottom) {
       el.scrollTo({ top: el.scrollHeight, behavior: "smooth" });
     }
-  }, [messages, autoScroll]);
+  }, [messages, streaming, autoScroll]);
 
   const submitConversation = useCallback(
     async (messagesToSend: ChatMessage[]) => {
       if (isLoading) return;
       const last = messagesToSend[messagesToSend.length - 1];
-      if (!last || last.role !== "user" || !last.content.trim()) return;
+      if (!last || last.role !== "user" || !hasText(last.content)) return;
 
       setError(null);
-      const assistantPlaceholder: ChatMessage = {
-        role: "assistant",
-        content: "",
-        toolCalls: undefined,
-      };
-      setMessages([...messagesToSend, assistantPlaceholder]);
+      setMessages(messagesToSend);
+      setStreaming({ content: "", reasoning: "", chain: [] });
       setIsLoading(true);
 
-      const apiMessages = messagesToSend.map((m) => ({
-        role: m.role,
-        content: m.content,
-      }));
-
       const baseURL = axios.defaults.baseURL ?? "";
       const url = `${baseURL}chat/completion`;
       const headers: Record<string, string> = {
@@ -104,16 +112,50 @@ export default function ChatPage() {
       const controller = new AbortController();
       abortRef.current = controller;
 
+      let chain: ChatMessage[] = [];
+      let stats: ChatStats | undefined;
+      let reasoning = "";
+      let hadError = false;
+
       await streamChatCompletion(
         url,
         headers,
-        apiMessages,
+        toWire(messagesToSend),
         {
-          updateMessages: (updater) => setMessages(updater),
-          onError: (message) => setError(message),
+          onContentDelta: (delta) =>
+            setStreaming((s) => (s ? { ...s, content: s.content + delta } : s)),
+          onReasoningDelta: (delta) => {
+            reasoning += delta;
+            setStreaming((s) =>
+              s ? { ...s, reasoning: s.reasoning + delta } : s,
+            );
+          },
+          onChain: (fullChain) => {
+            chain = fullChain;
+            setStreaming((s) => (s ? { ...s, chain: fullChain } : s));
+          },
+          onStats: (s) => {
+            stats = s;
+            setStreaming((cur) => (cur ? { ...cur, stats: s } : cur));
+          },
+          onError: (message) => {
+            hadError = true;
+            setError(message);
+          },
           onDone: () => {
             abortRef.current = null;
             setIsLoading(false);
+            setStreaming(null);
+            const lastMsg = chain[chain.length - 1];
+            if (!hadError && lastMsg?.role === "assistant") {
+              setMessages(
+                chain.map((m, i) =>
+                  i === chain.length - 1
+                    ? { ...m, reasoning: reasoning || undefined, stats }
+                    : m,
+                ),
+              );
+            }
           },
           defaultErrorMessage: t("error"),
         },
@@ -125,12 +167,14 @@ export default function ChatPage() {
   );
 
   const recentEventIds = useMemo(() => {
+    const responses = toolResponsesById(messages);
     for (let i = messages.length - 1; i >= 0; i--) {
       const msg = messages[i];
-      if (msg.role !== "assistant" || !msg.toolCalls) continue;
-      const similar = getFindSimilarObjectsFromToolCalls(msg.toolCalls);
+      if (msg.role !== "assistant" || !msg.tool_calls?.length) continue;
+      const calls = toolCallsForMessage(msg, responses);
+      const similar = getFindSimilarObjectsFromToolCalls(calls);
       if (similar) return similar.results.map((e) => e.id);
-      const events = getEventIdsFromSearchObjectsToolCalls(msg.toolCalls);
+      const events = getEventIdsFromSearchObjectsToolCalls(calls);
       if (events.length > 0) return events.map((e) => e.id);
     }
     return [];
@@ -154,12 +198,14 @@ export default function ChatPage() {
     abortRef.current?.abort();
     abortRef.current = null;
     setIsLoading(false);
+    setStreaming(null);
   }, []);
 
   const startNewChat = useCallback(() => {
     abortRef.current?.abort();
     abortRef.current = null;
     setIsLoading(false);
+    setStreaming(null);
     setMessages([]);
     setInput("");
     setAttachedEventId(null);
@@ -181,7 +227,83 @@ export default function ChatPage() {
     setAttachedEventId(null);
   }, []);
 
-  const hasStarted = messages.length > 0;
+  const hasStarted = messages.length > 0 || streaming != null;
+
+  // While streaming, the backend's in-flight chain is the source of truth;
+  // otherwise the committed conversation is.
+  const renderList =
+    streaming && streaming.chain.length ? streaming.chain : messages;
+  const responses = toolResponsesById(renderList);
+  const renderTail = renderList[renderList.length - 1];
+  const finalShown =
+    renderTail?.role === "assistant" && hasText(renderTail.content);
+
+  const renderMessage = (msg: ChatMessage, i: number) => {
+    if (msg.role === "system" || msg.role === "tool") return null;
+
+    if (msg.role === "user") {
+      if (!hasText(msg.content)) return null;
+      return (
+        <div key={i} className="flex flex-col gap-2">
+          <MessageBubble
+            role="user"
+            content={msg.content}
+            messageIndex={i}
+            onEditSubmit={handleEditSubmit}
+            isComplete
+            showStats={showStats}
+          />
+        </div>
+      );
+    }
+
+    const calls = toolCallsForMessage(msg, responses);
+    const contentText = hasText(msg.content) ? msg.content : "";
+    const similar = getFindSimilarObjectsFromToolCalls(calls);
+    const events = similar ? [] : getEventIdsFromSearchObjectsToolCalls(calls);
+
+    return (
+      <div key={i} className="flex flex-col gap-2">
+        {calls.length > 0 && <ToolCallsGroup toolCalls={calls} />}
+        {hasText(msg.reasoning) && (
+          <ReasoningBubble
+            reasoning={msg.reasoning}
+            answerStarted={!!contentText}
+          />
+        )}
+        {contentText && (
+          <MessageBubble
+            role="assistant"
+            content={contentText}
+            messageIndex={i}
+            isComplete
+            stats={msg.stats}
+            showStats={showStats}
+          />
+        )}
+        {similar ? (
+          <ChatEventThumbnailsRow
+            events={similar.results}
+            anchor={similar.anchor}
+            onAttach={setAttachedEventId}
+          />
+        ) : (
+          <ChatEventThumbnailsRow
+            events={events}
+            onAttach={setAttachedEventId}
+          />
+        )}
+      </div>
+    );
+  };
+
+  const processingDots = (
+    <div className="flex items-center gap-2 self-start rounded-2xl bg-muted px-5 py-4">
+      <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.32s]" />
+      <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.16s]" />
+      <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60" />
+    </div>
+  );
 
   return (
     <div className="flex size-full flex-col">
@@ -212,102 +334,31 @@ export default function ChatPage() {
           <div className="flex w-full flex-col xl:w-[50%] 3xl:w-[35%]">
             {hasStarted ? (
               <div className="flex w-full flex-1 flex-col gap-3 pb-3">
-                {messages.map((msg, i) => {
-                  const isLastAssistant =
-                    i === messages.length - 1 && msg.role === "assistant";
-                  const isComplete =
-                    msg.role === "user" || !isLoading || !isLastAssistant;
-                  const hasToolCalls =
-                    msg.toolCalls && msg.toolCalls.length > 0;
-                  const hasContent = !!msg.content?.trim();
-                  const hasReasoning = !!msg.reasoning?.trim();
-                  const showProcessing =
-                    isLastAssistant &&
-                    isLoading &&
-                    !hasContent &&
-                    !hasReasoning;
-
-                  // Hide empty placeholder only when there are no tool calls
-                  // and no reasoning streaming yet
-                  if (
-                    isLastAssistant &&
-                    isLoading &&
-                    !hasContent &&
-                    !hasToolCalls &&
-                    !hasReasoning
-                  )
-                    return (
-                      <div
-                        key={i}
-                        className="flex items-center gap-2 self-start rounded-2xl bg-muted px-5 py-4"
-                      >
-                        <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.32s]" />
-                        <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.16s]" />
-                        <span className="size-2.5 animate-bounce rounded-full bg-muted-foreground/60" />
-                      </div>
-                    );
-
-                  return (
-                    <div key={i} className="flex flex-col gap-2">
-                      {msg.role === "assistant" && hasToolCalls && (
-                        <ToolCallsGroup toolCalls={msg.toolCalls!} />
-                      )}
-                      {msg.role === "assistant" && hasReasoning && (
+                {renderList.map((msg, i) => renderMessage(msg, i))}
+                {streaming &&
+                  !finalShown &&
+                  (streaming.content || streaming.reasoning ? (
+                    <div className="flex flex-col gap-2">
+                      {hasText(streaming.reasoning) && (
                         <ReasoningBubble
-                          reasoning={msg.reasoning!}
-                          answerStarted={hasContent}
+                          reasoning={streaming.reasoning}
+                          answerStarted={!!streaming.content}
                         />
                       )}
-                      {showProcessing ? (
-                        <div className="flex items-center gap-2 self-start rounded-2xl bg-muted px-5 py-4">
-                          <span className="size-2 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.3s]" />
-                          <span className="size-2 animate-bounce rounded-full bg-muted-foreground/60 [animation-delay:-0.15s]" />
-                          <span className="size-2 animate-bounce rounded-full bg-muted-foreground/60" />
-                        </div>
-                      ) : msg.role === "assistant" &&
-                        !hasContent &&
-                        hasReasoning &&
-                        !isComplete ? null : (
+                      {streaming.content && (
                         <MessageBubble
-                          role={msg.role}
-                          content={msg.content}
-                          messageIndex={i}
-                          onEditSubmit={
-                            msg.role === "user" ? handleEditSubmit : undefined
-                          }
-                          isComplete={isComplete}
-                          stats={msg.stats}
+                          role="assistant"
+                          content={streaming.content}
+                          messageIndex={-1}
+                          isComplete={false}
+                          stats={streaming.stats}
                           showStats={showStats}
                         />
                       )}
-                      {msg.role === "assistant" &&
-                        isComplete &&
-                        (() => {
-                          const similar = getFindSimilarObjectsFromToolCalls(
-                            msg.toolCalls,
-                          );
-                          if (similar) {
-                            return (
-                              <ChatEventThumbnailsRow
-                                events={similar.results}
-                                anchor={similar.anchor}
-                                onAttach={setAttachedEventId}
-                              />
-                            );
-                          }
-                          const events = getEventIdsFromSearchObjectsToolCalls(
-                            msg.toolCalls,
-                          );
-                          return (
-                            <ChatEventThumbnailsRow
-                              events={events}
-                              onAttach={setAttachedEventId}
-                            />
-                          );
-                        })()}
                     </div>
-                  );
-                })}
+                  ) : (
+                    processingDots
+                  ))}
                 {error && (
                   <p
                     className="flex items-center gap-1.5 self-start text-sm text-destructive"

web/src/types/chat.ts

@@ -1,17 +1,30 @@
+export type ToolCallFunction = {
+  name: string;
+  arguments: string;
+};
+
+export type WireToolCall = {
+  id: string;
+  type?: string;
+  function: ToolCallFunction;
+};
+
+export type ChatMessage = {
+  role: "system" | "user" | "assistant" | "tool";
+  content: unknown;
+  tool_call_id?: string;
+  name?: string;
+  tool_calls?: WireToolCall[];
+  reasoning?: string;
+  stats?: ChatStats;
+};
+
 export type ToolCall = {
   name: string;
   arguments?: Record<string, unknown>;
   response?: string;
 };
 
-export type ChatMessage = {
-  role: "user" | "assistant";
-  content: string;
-  reasoning?: string;
-  toolCalls?: ToolCall[];
-  stats?: ChatStats;
-};
-
 export type StartingRequest = {
   label: string;
   prompt: string;

web/src/types/stats.ts

@@ -135,3 +135,22 @@ export type Ffprobe = {
     }[];
   };
 };
+
+export type KeyframeSeverity =
+  | "ok"
+  | "warning"
+  | "error"
+  | "unknown"
+  | "record_disabled";
+
+export type KeyframeAnalysis = {
+  severity: KeyframeSeverity;
+  stream_index?: number;
+  keyframe_count?: number;
+  max_gap?: number | null;
+  mean_gap?: number | null;
+  min_gap?: number | null;
+  duration_observed?: number | null;
+  segment_time?: number;
+  thresholds?: { warning: number; error: number };
+};

web/src/utils/chatUtil.ts

@@ -1,16 +1,20 @@
 import type { ChatMessage, ChatStats, ToolCall } from "@/types/chat";
 
 export type StreamChatCallbacks = {
-  /** Update the messages array (e.g. pass to setState). */
-  updateMessages: (updater: (prev: ChatMessage[]) => ChatMessage[]) => void;
+  /** Streamed delta of the assistant's final answer text. */
+  onContentDelta: (delta: string) => void;
+  /** Streamed delta of the assistant's reasoning trace. */
+  onReasoningDelta: (delta: string) => void;
+  /** The full conversation chain so far (system message, history, this turn's
+   * tool-call turns, tool results, and — on the final emission — the final
+   * assistant message). */
+  onChain: (chain: ChatMessage[]) => void;
+  /** Token/timing stats for the turn. */
+  onStats: (stats: ChatStats) => void;
   /** Called when the stream sends an error or fetch fails. */
   onError: (message: string) => void;
   /** Called when the stream finishes (success or error). */
   onDone: () => void;
-  /** Called when the stream emits token/timing stats. The stats are also
-   * attached to the last assistant message in updateMessages, so consumers
-   * can usually rely on the message itself rather than wiring this up. */
-  onStats?: (stats: ChatStats) => void;
   /** Message used when fetch throws and no server error is available. */
   defaultErrorMessage?: string;
 };
@@ -25,7 +29,7 @@ type StatsChunk = {
 
 type StreamChunk =
   | { type: "error"; error: string }
-  | { type: "tool_calls"; tool_calls: ToolCall[] }
+  | { type: "messages"; messages: ChatMessage[] }
   | { type: "content"; delta: string }
   | { type: "reasoning"; delta: string }
   | StatsChunk;
@@ -41,16 +45,18 @@ export type StreamChatOptions = {
 export async function streamChatCompletion(
   url: string,
   headers: Record<string, string>,
-  apiMessages: { role: string; content: string }[],
+  apiMessages: ChatMessage[],
   callbacks: StreamChatCallbacks,
   signal?: AbortSignal,
   options: StreamChatOptions = {},
 ): Promise<void> {
   const {
-    updateMessages,
+    onContentDelta,
+    onReasoningDelta,
+    onChain,
+    onStats,
     onError,
     onDone,
-    onStats,
     defaultErrorMessage = "Something went wrong. Please try again.",
   } = callbacks;
 
@@ -91,65 +97,27 @@ export async function streamChatCompletion(
     const applyChunk = (data: StreamChunk) => {
       if (data.type === "error") {
         onError(data.error);
-        updateMessages((prev) =>
-          prev.filter((m) => !(m.role === "assistant" && m.content === "")),
-        );
         return "break";
       }
-      if (data.type === "tool_calls" && data.tool_calls?.length) {
-        updateMessages((prev) => {
-          const next = [...prev];
-          const lastMsg = next[next.length - 1];
-          if (lastMsg?.role === "assistant")
-            next[next.length - 1] = {
-              ...lastMsg,
-              toolCalls: data.tool_calls,
-            };
-          return next;
-        });
+      if (data.type === "messages") {
+        onChain(data.messages ?? []);
         return "continue";
       }
       if (data.type === "content" && data.delta !== undefined) {
-        updateMessages((prev) => {
-          const next = [...prev];
-          const lastMsg = next[next.length - 1];
-          if (lastMsg?.role === "assistant")
-            next[next.length - 1] = {
-              ...lastMsg,
-              content: lastMsg.content + data.delta,
-            };
-          return next;
-        });
+        onContentDelta(data.delta);
         return "continue";
       }
       if (data.type === "reasoning" && data.delta !== undefined) {
-        updateMessages((prev) => {
-          const next = [...prev];
-          const lastMsg = next[next.length - 1];
-          if (lastMsg?.role === "assistant")
-            next[next.length - 1] = {
-              ...lastMsg,
-              reasoning: (lastMsg.reasoning ?? "") + data.delta,
-            };
-          return next;
-        });
+        onReasoningDelta(data.delta);
         return "continue";
       }
       if (data.type === "stats") {
-        const stats: ChatStats = {
+        onStats({
           promptTokens: data.prompt_tokens,
           completionTokens: data.completion_tokens,
           completionDurationMs: data.completion_duration_ms,
           tokensPerSecond: data.tokens_per_second,
-        };
-        updateMessages((prev) => {
-          const next = [...prev];
-          const lastMsg = next[next.length - 1];
-          if (lastMsg?.role === "assistant")
-            next[next.length - 1] = { ...lastMsg, stats };
-          return next;
         });
-        onStats?.(stats);
         return "continue";
       }
       return "continue";
@@ -165,9 +133,8 @@ export async function streamChatCompletion(
         const trimmed = line.trim();
         if (!trimmed) continue;
         try {
-          const data = JSON.parse(trimmed) as StreamChunk & { type: string };
-          const result = applyChunk(data as StreamChunk);
-          if (result === "break") {
+          const data = JSON.parse(trimmed) as StreamChunk;
+          if (applyChunk(data) === "break") {
             hadStreamError = true;
             break;
           }
@@ -181,50 +148,63 @@ export async function streamChatCompletion(
     // Flush remaining buffer
     if (!hadStreamError && buffer.trim()) {
       try {
-        const data = JSON.parse(buffer.trim()) as StreamChunk & {
-          type: string;
-          delta?: string;
-        };
-        if (data.type === "content" && data.delta !== undefined) {
-          updateMessages((prev) => {
-            const next = [...prev];
-            const lastMsg = next[next.length - 1];
-            if (lastMsg?.role === "assistant")
-              next[next.length - 1] = {
-                ...lastMsg,
-                content: lastMsg.content + data.delta!,
-              };
-            return next;
-          });
-        }
+        const data = JSON.parse(buffer.trim()) as StreamChunk;
+        applyChunk(data);
       } catch {
         // ignore final malformed chunk
       }
     }
-
-    if (!hadStreamError) {
-      updateMessages((prev) => {
-        const next = [...prev];
-        const lastMsg = next[next.length - 1];
-        if (lastMsg?.role === "assistant" && lastMsg.content === "")
-          next[next.length - 1] = { ...lastMsg, content: " " };
-        return next;
-      });
-    }
   } catch (err) {
     if (err instanceof DOMException && err.name === "AbortError") {
       // User stopped generation — not an error
     } else {
       onError(defaultErrorMessage);
-      updateMessages((prev) =>
-        prev.filter((m) => !(m.role === "assistant" && m.content === "")),
-      );
     }
   } finally {
     onDone();
   }
 }
 
+/** Map each tool result message to its tool_call_id for response lookup. */
+export function toolResponsesById(
+  messages: ChatMessage[],
+): Map<string, string> {
+  const map = new Map<string, string>();
+  for (const m of messages) {
+    if (m.role === "tool" && typeof m.tool_call_id === "string") {
+      map.set(
+        m.tool_call_id,
+        typeof m.content === "string" ? m.content : JSON.stringify(m.content),
+      );
+    }
+  }
+  return map;
+}
+
+/** Derive the display tool calls for one assistant message. */
+export function toolCallsForMessage(
+  message: ChatMessage,
+  responses: Map<string, string>,
+): ToolCall[] {
+  if (!message.tool_calls?.length) return [];
+  return message.tool_calls.map((tc) => {
+    let args: Record<string, unknown> | undefined;
+    const raw = tc.function?.arguments;
+    if (typeof raw === "string") {
+      try {
+        args = JSON.parse(raw) as Record<string, unknown>;
+      } catch {
+        args = undefined;
+      }
+    }
+    return {
+      name: tc.function?.name ?? "",
+      arguments: args,
+      response: responses.get(tc.id),
+    };
+  });
+}
+
 /**
  * Parse search_objects tool call response(s) into event ids for thumbnails.
  */