wip introduce a performance monitor in-app

The Kyutai open-source model turned out not to be production-ready:
it caused disruptions in the production environment, especially on
long-running meeting sessions.

Switch to the Voxtral realtime model, which looks like a
credible competitor and behaves much better in our setup.

For now, the code handling the Voxtral realtime API lives directly
in the project. It could be extracted into an open-source package
later.

See PR #1277 for the full details of the implementation, proposed
by @cameldev.

.gitignore

@@ -75,7 +75,6 @@ db.sqlite3
 # IDEs
 .idea/
 .vscode/
-.cursor/
 *.iml
 .devcontainer
 

CHANGELOG.md

@@ -10,7 +10,12 @@ and this project adheres to
 
 ### Added
 
-- ✨(frontend) share OneToOneFocusLayout between PiP and main room
+- ✨(agent) support Voxtral realtime as inference engine
+
+### Changed
+
+- 🔥(backend) remove the S3 storage-event webhook for recordings
+- ♻️(backend) always finalize recordings using the LiveKit egress_ended webhook
 
 ## [1.29.0] - 2026-08-25
 

UPGRADE.md

@@ -16,6 +16,23 @@ the following command inside your docker container:
 
 ## [Unreleased]
 
+### Removing S3 storage-event webhooks for recordings
+
+Recordings were previously confirmed as saved by an S3 storage-event webhook posting to `/api/v1.0/recordings/storage-hook/`. That endpoint has been removed: recordings are now always finalized from LiveKit's own `egress_ended` webhook, which has been the default path since v1.22.0.
+
+**Required for every deployment:** LiveKit must be able to deliver webhooks to the backend at `/api/v1.0/rooms/webhooks-livekit/`. This is now the only way a recording reaches a saved state; if `egress_ended` is never delivered, recordings stay in the `active` state.
+
+For hosters who had configured storage-event webhooks:
+- Recordings reach the same final state, but they are now finalized when LiveKit reports the egress as ended rather than when the storage backend reports the upload.
+- Remove the event notification from your bucket configuration: it now targets a non-existent endpoint and will fail on every delivery.
+
+For hosters who had **not** configured storage-event webhooks:
+- Nothing changes. Recordings have been finalized from the `egress_ended` webhook since v1.22.0.
+
+In both cases, the following settings are no longer used and can be removed from your env: `RECORDING_EVENT_PARSER_CLASS`, `RECORDING_ENABLE_STORAGE_EVENT_AUTH`, `RECORDING_STORAGE_EVENT_ENABLE`, `RECORDING_STORAGE_EVENT_TOKEN`.
+
+On completion of the egress, a recording moves to `notification_succeeded`, or to `saved` if notifying external services failed.
+
 ## v1.23.0
 
 As part of the 1.23.0 release, the legacy `api/v1` implementation has been removed from the _experimental_ Summary service and Meet has been migrated to the new `api/v2`.

compose.yml

@@ -46,21 +46,6 @@ services:
       /usr/bin/mc mb meet/meet-media-storage && \
       exit 0;"
 
-  createwebhook:
-    image: minio/mc
-    depends_on:
-      minio:
-        condition: service_healthy
-        restart: true
-    entrypoint: >
-      sh -c "
-      /usr/bin/mc alias set meet http://minio:9000 meet password &&
-      /usr/bin/mc admin config set meet notify_webhook:meet-webhook endpoint='http://app-dev:8000/api/v1.0/recordings/storage-hook/' auth_token='Bearer password' &&
-      /usr/bin/mc admin service restart meet --wait --json &&
-      sleep 15 &&
-      /usr/bin/mc event add meet/meet-media-storage arn:minio:sqs::meet-webhook:webhook --event put --prefix "recordings" &&
-      exit 0;"
-
   app-dev:
     build:
       context: .
@@ -86,7 +71,6 @@ services:
         - mailcatcher
         - redis
         - createbuckets
-        - createwebhook
     extra_hosts:
       - "127.0.0.1.nip.io:host-gateway"
     networks:
@@ -229,6 +213,8 @@ services:
       - "7880:7880"
       - "7881:7881"
       - "7882:7882/udp"
+      - "3478:3478/udp"
+      - "30000-30100:30000-30100/udp"
     volumes:
       - ./docker/livekit/config/livekit-server.yaml:/config.yaml
     depends_on:
@@ -267,6 +253,7 @@ services:
     build:
       context: ./src/agents
       target: development
+    command: ["python", "multi_user_transcriber.py", "dev"]
     env_file:
       - env.d/development/multi_user_transcriber
     volumes:

docker/livekit/config/livekit-server.yaml

@@ -8,3 +8,16 @@ webhook:
   api_key: devkey
   urls:
       - http://app-dev:8000/api/v1.0/rooms/webhooks-livekit/
+
+turn:
+  enabled: true
+  domain: turn.192.168.1.50.nip.io
+  udp_port: 3478
+  tls_port: 0
+  external_tls: false
+  relay_range_start: 30000
+  relay_range_end: 30100
+  allow_restricted_peer_cidrs:
+    - 192.168.0.0/16
+    - 172.16.0.0/12
+

docs/features/recording.md

@@ -23,14 +23,11 @@ It uses LiveKit Egress to record room sessions. For reference, see the [LiveKit
 To use the room recording feature, the following components are required:
 
 - A running [LiveKit Egress](https://github.com/livekit/egress) server capable of handling room composite recordings.
-- A S3-compatible object storage that supports webhook events to notify the backend when recordings are uploaded.
+- A S3-compatible object storage where the egress uploads the recorded files.
 - An email service to notify room owners when a recording is available for download.
 - Webhook events configured between LiveKit Server and the backend.
 
 
-> [!CAUTION]
->  Minio supports lifecycle events; other providers may not work out of the box. There is currently a dependency on Minio, which is planned to be refactored in the future.
-
 > [!NOTE]
 > Celery isn’t in use for these async tasks yet. It’s something we’d like to add, but it’s not planned at this stage.
 
@@ -75,7 +72,7 @@ sequenceDiagram
     LiveKit->>Egress: Stop recording
     Egress->>Storage: Upload recorded file
     
-    Storage->>Backend: Storage event notification
+    LiveKit->>Backend: POST /api/v1.0/rooms/webhooks-livekit/ (egress_ended)
     Backend->>Backend: Update Recording status to SAVED
     Backend->>Email: Send notification to room owner
     
@@ -94,10 +91,6 @@ sequenceDiagram
 | **RECORDING_ENABLE**                    | Boolean     | `False`                                                                                                                                                            | Enable or disable the room recording feature.                                                                                                                                                                                                                                                      |
 | **RECORDING_OUTPUT_FOLDER**             | String      | `"recordings"`                                                                                                                                                     | Folder/prefix where recordings are stored in the object storage.                                                                                                                                                                                                                                   |
 | **RECORDING_WORKER_CLASSES**            | Dict        | `{ "screen_recording": "core.recording.worker.services.VideoCompositeEgressService", "transcript": "core.recording.worker.services.AudioCompositeEgressService" }` | Maps recording types to their worker service classes.                                                                                                                                                                                                                                              |
-| **RECORDING_EVENT_PARSER_CLASS**        | String      | `"core.recording.event.parsers.MinioParser"`                                                                                                                       | Class responsible for parsing storage events and updating the backend.                                                                                                                                                                                                                             |
-| **RECORDING_ENABLE_STORAGE_EVENT_AUTH** | Boolean     | `True`                                                                                                                                                             | Enable authentication for storage event webhook requests.                                                                                                                                                                                                                                          |
-| **RECORDING_STORAGE_EVENT_ENABLE**      | Boolean     | `False`                                                                                                                                                            | Enable handling of storage events (must configure webhook in storage). If `False`, fallback to LiveKit egress complete webhook.                                                    |
-| **RECORDING_STORAGE_EVENT_TOKEN**       | Secret/File | `None`                                                                                                                                                             | Token used to authenticate storage webhook requests, if `RECORDING_ENABLE_STORAGE_EVENT_AUTH` is enabled.                                                                                                                                                                                          |
 | **RECORDING_EXPIRATION_DAYS**           | Integer     | `None`                                                                                                                                                             | Number of days before recordings expire. Should match bucket lifecycle policy. Set to `None` for no expiration.                                                                                                                                                                                    |
 | **RECORDING_MAX_DURATION**              | Integer     | `None`                                                                                                                                                             | Maximum duration of a recording in milliseconds. Must be synced with the LiveKit Egress configuration. Set to None for unlimited duration. When the maximum duration is reached, the recording is automatically stopped and saved, and the user is prompted in the frontend with an alert message. |
 | **RECORDING_ENCODING_ENABLED**          | Boolean     | `False`                                                                                                                                                            | When `False`, LiveKit Egress uses its built-in `H264_720P_30` preset. When `True`, the `RECORDING_ENCODING_*` values below are sent to LiveKit as advanced `EncodingOptions`. See [Tuning recording encoding](#tuning-recording-encoding).                                                          |
@@ -109,19 +102,6 @@ sequenceDiagram
 | **RECORDING_ENCODING_KEY_FRAME_INTERVAL_S** | Float   | `4.0`                                                                                                                                                              | Keyframe interval in seconds. Drives seek granularity in the recorded MP4 (a player can only seek to keyframe boundaries). Larger values give the encoder slightly more bits for non-keyframe content at a fixed bitrate. `4.0` is a standard VOD value. Only applied when `RECORDING_ENCODING_ENABLED` is `True`. |
 
 
-### Manual Storage Webhook
-
-Storage events must be configured manually; the Kubernetes chart does not do this automatically.
-
-1. Configure your S3 bucket to send file creation events to the backend webhook.
-2. Enable events and token in settings:
-
-```python
-RECORDING_STORAGE_EVENT_ENABLE = True
-RECORDING_ENABLE_STORAGE_EVENT_AUTH = True
-RECORDING_STORAGE_EVENT_TOKEN = <token>
-```
-
 > [!NOTE]
 > Questions? Open an issue on [GitHub](https://github.com/suitenumerique/meet/issues/new?assignees=&labels=bug&template=Bug_report.md) or join our [Matrix community](https://matrix.to/#/#meet-official:matrix.org).
 

docs/installation/kubernetes.md

@@ -406,10 +406,6 @@ These are the environmental options available on meet backend.
 | RECORDING_ENABLE                                | Record meeting option                                                                                                                                        | false                                                                                                                                                         |
 | RECORDING_OUTPUT_FOLDER                         | Folder to store meetings                                                                                                                                     | recordings                                                                                                                                                    |
 | RECORDING_WORKER_CLASSES                        | Worker classes for recording                                                                                                                                 | {"screen_recording": "core.recording.worker.services.VideoCompositeEgressService","transcript": "core.recording.worker.services.AudioCompositeEgressService"} |
-| RECORDING_EVENT_PARSER_CLASS                    | Storage event engine for recording                                                                                                                           | core.recording.event.parsers.MinioParser                                                                                                                      |
-| RECORDING_ENABLE_STORAGE_EVENT_AUTH             | Enable storage event authorization                                                                                                                           | true                                                                                                                                                          |
-| RECORDING_STORAGE_EVENT_ENABLE                  | Enable recording storage events. If false, fallback to egress webhook.                                                                                                                           | false                                                                                                                                                         |
-| RECORDING_STORAGE_EVENT_TOKEN                   | Recording storage event token                                                                                                                                |                                                                                                                                                               |
 | RECORDING_EXPIRATION_DAYS                       | Recording expiration in days                                                                                                                                 |                                                                                                                                                               |
 | RECORDING_MAX_DURATION                          | Maximum recording duration in milliseconds. Must match LiveKit Egress configuration exactly.                                                                 |                                                                                                                                                               |
 | SCREEN_RECORDING_BASE_URL                       | Screen recording base URL                                                                                                                                    |                                                                                                                                                               |

env.d/development/common.dist

@@ -63,8 +63,6 @@ ALLOW_UNREGISTERED_ROOMS=False
 
 # Recording
 RECORDING_ENABLE=True
-RECORDING_STORAGE_EVENT_ENABLE=False
-RECORDING_STORAGE_EVENT_TOKEN=password
 SUMMARY_SERVICE_ENDPOINT=http://app-summary-dev:8000/api/v2/async-jobs/transcribe/
 SUMMARY_SERVICE_API_TOKEN=password
 SUMMARY_SERVICE_WEBHOOK_API_TOKEN=webhook-password

env.d/development/multi_user_transcriber.dist

@@ -1,14 +1,23 @@
+AWS_S3_ENDPOINT_URL=minio:9000
+AWS_S3_ACCESS_KEY_ID=meet
+AWS_S3_SECRET_ACCESS_KEY=password
+
 LIVEKIT_URL=ws://livekit:7880
 LIVEKIT_API_KEY=devkey
 LIVEKIT_API_SECRET=secret
 
-STT_PROVIDER=kyutai # kyutai, deepgram
+STT_PROVIDER=voxtral-vllm # voxtral-vllm, kyutai, deepgram
 ENABLE_SILERO_VAD=False
 
-DEEPGRAM_API_KEY=
+DEEPGRAM_API_KEY=your-deepgram-api-key
 
-KYUTAI_STT_BASE_URL=
-KYUTAI_API_KEY=
+KYUTAI_STT_BASE_URL=url
+KYUTAI_API_KEY=your-kyutai-api-key
+
+VOXTRAL_VLLM_BASE_URL=wss://<host>/v1/realtime
+VOXTRAL_VLLM_MODEL=voxtral-mini-4b-realtime-2602
+VOXTRAL_VLLM_API_KEY=your-vllm-api-key
+VOXTRAL_VLLM_TARGET_STREAMING_DELAY_MS=480
 
 SENTRY_DSN=
 SENTRY_ENVIRONMENT=

src/agents/multi_user_transcriber.py

@@ -1,6 +1,7 @@
 """Multi user transcription agent."""
 
 import asyncio
+import contextlib
 import logging
 import os
 
@@ -25,6 +26,7 @@ from livekit.agents import (
 )
 from livekit.plugins import deepgram, silero
 
+import voxtral_vllm_stt
 from observability import configure_sentry, set_job_context
 from tasks import done_callback
 
@@ -36,9 +38,18 @@ TRANSCRIBER_AGENT_NAME = os.getenv("TRANSCRIBER_AGENT_NAME", "multi-user-transcr
 STT_PROVIDER = os.getenv("STT_PROVIDER", "deepgram")
 ENABLE_SILERO_VAD = os.getenv("ENABLE_SILERO_VAD", "true").lower() == "true"
 
+SESSION_DRAIN_TIMEOUT_S = 15.0
 
-def create_stt_provider():
-    """Create STT provider based on environment configuration."""
+
+def create_stt_provider(vad: silero.VAD | None = None):
+    """Create STT provider based on environment configuration.
+
+    Args:
+        vad: Shared, prewarmed VAD instance. Required in practice for
+            voxtral-vllm (no server-side endpointing): if omitted, the plugin
+            loads its own Silero model synchronously on the event loop, once
+            per participant, freezing all active sessions for the duration.
+    """
     if STT_PROVIDER == "deepgram":
         # Note: Not all Deepgram API parameters are supported by the LiveKit plugin
         # detect_language is NOT supported for real-time streaming
@@ -49,6 +60,9 @@ def create_stt_provider():
         )
     elif STT_PROVIDER == "kyutai":
         _stt_instance = kyutai.STT(base_url=os.getenv("KYUTAI_STT_BASE_URL"))
+    elif STT_PROVIDER == "voxtral-vllm":
+        # The plugin resolves base_url / model / api_key from the environment.
+        _stt_instance = voxtral_vllm_stt.STT(vad=vad)
     else:
         raise ValueError(f"Unknown STT_PROVIDER: {STT_PROVIDER}")
 
@@ -58,9 +72,9 @@ def create_stt_provider():
 class Transcriber(Agent):
     """Create a transcription agent for a specific participant."""
 
-    def __init__(self, *, participant_identity: str):
+    def __init__(self, *, participant_identity: str, vad: silero.VAD | None = None):
         """Init transcription agent."""
-        stt = create_stt_provider()
+        stt = create_stt_provider(vad=vad)
 
         super().__init__(
             instructions="not-needed",
@@ -76,6 +90,7 @@ class MultiUserTranscriber:
         """Init multi user transcription agent."""
         self.ctx = ctx
         self._sessions: dict[str, AgentSession] = {}
+        self._starting: dict[str, asyncio.Task] = {}
         self._tasks: set[asyncio.Task] = set()
 
     def start(self):
@@ -96,22 +111,30 @@ class MultiUserTranscriber:
 
     def on_participant_connected(self, participant: rtc.RemoteParticipant):
         """Handle new participant connection by starting transcription session."""
-        if participant.identity in self._sessions:
+        identity = participant.identity
+        if identity in self._sessions or identity in self._starting:
             return
 
-        logger.info(f"starting session for {participant.identity}")
+        logger.info(f"starting session for {identity}")
         task = asyncio.create_task(self._start_session(participant))
+        self._starting[identity] = task
         self._tasks.add(task)
+        task.add_done_callback(lambda t, i=identity: self._starting.pop(i, None))
         task.add_done_callback(
             done_callback(
                 logger,
                 self._tasks,
-                f"start transcription session for {participant.identity}",
+                f"start transcription session for {identity}",
             )
         )
 
     def on_participant_disconnected(self, participant: rtc.RemoteParticipant):
         """Handle participant disconnection by closing transcription session."""
+        if (start_task := self._starting.pop(participant.identity, None)) is not None:
+            logger.info(f"cancelling pending session start for {participant.identity}")
+            start_task.cancel()
+            return
+
         if (session := self._sessions.pop(participant.identity, None)) is None:
             return
 
@@ -127,10 +150,12 @@ class MultiUserTranscriber:
         )
 
     async def _start_session(self, participant: rtc.RemoteParticipant) -> AgentSession:
-        """Create and start transcription session for participant."""
-        if participant.identity in self._sessions:
-            return self._sessions[participant.identity]
+        """Create and start transcription session for participant.
 
+        Deduplication happens synchronously in on_participant_connected via
+        self._starting; by the time this coroutine runs, the identity is
+        already reserved.
+        """
         vad = self.ctx.proc.userdata.get("vad", None)
         session = AgentSession(vad=vad)
         room_io = RoomIO(
@@ -141,18 +166,30 @@ class MultiUserTranscriber:
                 text_input=False, audio_output=False, text_output=True
             ),
         )
-        await room_io.start()
-        await session.start(
-            agent=Transcriber(
-                participant_identity=participant.identity,
+        try:
+            await room_io.start()
+            await session.start(
+                agent=Transcriber(
+                    participant_identity=participant.identity,
+                    vad=vad,
+                )
             )
-        )
+        except BaseException:
+            with contextlib.suppress(Exception):
+                await session.aclose()
+            raise
         self._sessions[participant.identity] = session
         return session
 
     async def _close_session(self, sess: AgentSession) -> None:
         """Close and cleanup transcription session."""
-        await sess.drain()
+        try:
+            await asyncio.wait_for(sess.drain(), timeout=SESSION_DRAIN_TIMEOUT_S)
+        except (TimeoutError, asyncio.TimeoutError):
+            logger.warning(
+                "session drain timed out after %.0fs; forcing close",
+                SESSION_DRAIN_TIMEOUT_S,
+            )
         await sess.aclose()
 
 

src/agents/pyproject.toml

@@ -12,6 +12,8 @@ dependencies = [
     "protobuf==6.33.6",
     "minio==7.2.20",
     "sentry-sdk==2.66.1",
+    "websockets==17.1",
+    "httpx==0.28.1",
 ]
 
 [project.optional-dependencies]

src/agents/voxtral_vllm_stt.py

@@ -0,0 +1,476 @@
+"""LiveKit STT plugin for Voxtral Realtime served via vLLM (/v1/realtime).
+
+vLLM exposes Voxtral Realtime over a WebSocket that follows the OpenAI Realtime
+API protocol (not Mistral's proprietary realtime protocol).
+"""
+
+from __future__ import annotations
+
+import asyncio
+import base64
+import json
+import logging
+import os
+import weakref
+from collections import deque
+from dataclasses import dataclass, field
+
+import websockets
+from livekit.agents import (
+    DEFAULT_API_CONNECT_OPTIONS,
+    APIConnectionError,
+    APIConnectOptions,
+    APIStatusError,
+    stt,
+    utils,
+)
+from livekit.agents import (
+    vad as vad_module,
+)
+from livekit.agents.types import NOT_GIVEN, NotGivenOr
+from livekit.agents.utils import is_given
+
+logger = logging.getLogger("voxtral-vllm-stt")
+
+SAMPLE_RATE = 16000
+NUM_CHANNELS = 1
+CHUNK_SAMPLES = 1600  # 100 ms @ 16 kHz mono
+PREROLL_CHUNKS = 5  # keep 500 ms of audio before start of speech as detected by VAD
+
+# Reconnect policy: exponential backoff capped at MAX, give up after MAX_ATTEMPTS
+# consecutive failures (a successful handshake resets the counter).
+RECONNECT_BACKOFF_BASE_S = 0.5
+RECONNECT_BACKOFF_MAX_S = 8.0
+RECONNECT_MAX_ATTEMPTS = 5
+
+
+@dataclass
+class _STTOptions:
+    base_url: str
+    model: str
+    api_key: str | None
+    target_streaming_delay_ms: int | None
+
+
+@dataclass
+class _PendingUtterance:
+    """An utterance in flight on the shared websocket used for reconnect.
+
+    `sent_chunks` holds every chunk we have already enqueued for send on this
+    or a prior connection; on reconnect we replay them before resuming reads
+    from `queue`. vLLM concatenates `input_audio_buffer.append` events into a
+    single audio buffer per generation, so duplicates from a partial prior send
+    are harmless.
+    """
+
+    queue: asyncio.Queue[bytes | None]
+    sent_chunks: list[bytes] = field(default_factory=list)
+    ended: bool = False
+
+
+class STT(stt.STT):
+    """LiveKit STT speaking the OpenAI Realtime protocol served by vLLM."""
+
+    def __init__(
+        self,
+        *,
+        base_url: NotGivenOr[str] = NOT_GIVEN,
+        model: NotGivenOr[str] = NOT_GIVEN,
+        api_key: NotGivenOr[str] = NOT_GIVEN,
+        target_streaming_delay_ms: NotGivenOr[int] = NOT_GIVEN,
+        vad: vad_module.VAD | None = None,
+    ) -> None:
+        """Build the STT.
+
+        Args:
+            base_url: WebSocket URL of the vLLM realtime endpoint, e.g.
+                ws://example:8000/v1/realtime. Falls back to $VOXTRAL_VLLM_BASE_URL.
+            model: Model name exposed by vLLM, default
+                mistralai/Voxtral-Mini-4B-Realtime-2602.
+            api_key: Optional bearer token. Falls back to $VOXTRAL_VLLM_API_KEY.
+            target_streaming_delay_ms: Target streaming delay in ms forwarded to
+                vLLM via session.update. Falls back to
+                $VOXTRAL_VLLM_TARGET_STREAMING_DELAY_MS, else server default.
+            vad: Voice Activity Detector. If omitted, Silero VAD is loaded.
+        """
+        super().__init__(
+            capabilities=stt.STTCapabilities(streaming=True, interim_results=True)
+        )
+
+        resolved_url = (
+            base_url
+            if is_given(base_url)
+            else os.environ.get(
+                "VOXTRAL_VLLM_BASE_URL", "ws://127.0.0.1:8000/v1/realtime"
+            )
+        )
+        resolved_model = (
+            model
+            if is_given(model)
+            else os.environ.get(
+                "VOXTRAL_VLLM_MODEL", "mistralai/Voxtral-Mini-4B-Realtime-2602"
+            )
+        )
+        resolved_key = (
+            api_key if is_given(api_key) else os.environ.get("VOXTRAL_VLLM_API_KEY")
+        )
+        resolved_delay = (
+            target_streaming_delay_ms
+            if is_given(target_streaming_delay_ms)
+            else (
+                int(os.environ["VOXTRAL_VLLM_TARGET_STREAMING_DELAY_MS"])
+                if os.environ.get("VOXTRAL_VLLM_TARGET_STREAMING_DELAY_MS")
+                else None
+            )
+        )
+
+        if vad is None:
+            try:
+                from livekit.plugins.silero import VAD as SileroVAD  # noqa: PLC0415
+            except ImportError as exc:
+                raise ImportError(
+                    "livekit-plugins-silero is required for vLLM Voxtral realtime "
+                    "(no server-side endpointing)."
+                ) from exc
+            vad = SileroVAD.load()
+        self._vad = vad
+
+        self._opts = _STTOptions(
+            base_url=resolved_url,
+            model=resolved_model,
+            api_key=resolved_key,
+            target_streaming_delay_ms=resolved_delay,
+        )
+        self._streams: weakref.WeakSet[SpeechStream] = weakref.WeakSet()
+
+    @property
+    def model(self) -> str:
+        """Return the configured vLLM model name."""
+        return self._opts.model
+
+    @property
+    def provider(self) -> str:
+        """Return the provider identifier."""
+        return "vllm-voxtral-realtime"
+
+    async def _recognize_impl(self, *_args, **_kwargs) -> stt.SpeechEvent:
+        raise NotImplementedError(
+            "vLLM Voxtral Realtime STT only supports streaming recognition."
+        )
+
+    def stream(
+        self,
+        *,
+        conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS,
+    ) -> SpeechStream:
+        """Open a new streaming recognition stream."""
+        s = SpeechStream(
+            stt=self,
+            opts=self._opts,
+            vad_instance=self._vad,
+            conn_options=conn_options,
+        )
+        self._streams.add(s)
+        return s
+
+
+class SpeechStream(stt.RecognizeStream):
+    """Voxtral realtime handler."""
+
+    def __init__(
+        self,
+        *,
+        stt: STT,
+        opts: _STTOptions,
+        vad_instance: vad_module.VAD,
+        conn_options: APIConnectOptions,
+    ) -> None:
+        """Init the speech stream."""
+        super().__init__(stt=stt, conn_options=conn_options, sample_rate=SAMPLE_RATE)
+        self._opts = opts
+        self._vad = vad_instance
+        self._utterance_q: asyncio.Queue[bytes | None] | None = None
+        self._speaking = False
+        self._preroll: deque[bytes] = deque(maxlen=PREROLL_CHUNKS)
+        # Voxtral realtime is strictly sequential: only one generation runs at a
+        # time, and a new `commit` is ignored while the previous one is still
+        # producing. We queue per-utterance audio buffers here and let the
+        # pipeline process them one by one on the shared websocket.
+        self._utterance_chan: asyncio.Queue[asyncio.Queue[bytes | None] | None] = (
+            asyncio.Queue()
+        )
+
+    @utils.log_exceptions(logger=logger)
+    async def _run(self) -> None:
+        vad_stream = self._vad.stream()
+
+        bstream = utils.audio.AudioByteStream(
+            sample_rate=SAMPLE_RATE,
+            num_channels=NUM_CHANNELS,
+            samples_per_channel=CHUNK_SAMPLES,
+        )
+
+        async def input_task() -> None:
+            async for data in self._input_ch:
+                if isinstance(data, self._FlushSentinel):
+                    for frame in bstream.flush():
+                        self._handle_chunk(frame.data.tobytes())
+                    continue
+
+                vad_stream.push_frame(data)
+                for frame in bstream.write(data.data.tobytes()):
+                    self._handle_chunk(frame.data.tobytes())
+
+            vad_stream.end_input()
+
+        async def vad_task() -> None:
+            async for ev in vad_stream:
+                if ev.type == vad_module.VADEventType.START_OF_SPEECH:
+                    self._on_start_of_speech()
+                elif ev.type == vad_module.VADEventType.END_OF_SPEECH:
+                    self._on_end_of_speech()
+
+        pipeline_t = asyncio.create_task(self._utterance_pipeline())
+        try:
+            await asyncio.gather(input_task(), vad_task())
+            # signal end-of-stream; pipeline finishes pending utterances first
+            self._utterance_chan.put_nowait(None)
+            await pipeline_t
+        except (APIStatusError, APIConnectionError, asyncio.CancelledError):
+            raise
+        except Exception as exc:
+            logger.exception("vLLM realtime stream failed")
+            raise APIConnectionError() from exc
+        finally:
+            if not pipeline_t.done():
+                pipeline_t.cancel()
+            try:
+                await pipeline_t
+            except asyncio.CancelledError:
+                # CancelledError is the expected flow on cancel()
+                pass
+            except Exception:
+                logger.exception("utterance pipeline failed during finalize")
+            await vad_stream.aclose()
+
+    def _handle_chunk(self, chunk: bytes) -> None:
+        self._preroll.append(chunk)
+        if self._speaking and self._utterance_q is not None:
+            self._utterance_q.put_nowait(chunk)
+
+    def _on_start_of_speech(self) -> None:
+        if self._speaking:
+            return
+        self._speaking = True
+        q: asyncio.Queue[bytes | None] = asyncio.Queue()
+        for chunk in self._preroll:
+            q.put_nowait(chunk)
+        self._utterance_q = q
+        self._utterance_chan.put_nowait(q)
+        self._event_ch.send_nowait(
+            stt.SpeechEvent(type=stt.SpeechEventType.START_OF_SPEECH)
+        )
+
+    def _on_end_of_speech(self) -> None:
+        if not self._speaking:
+            return
+        self._speaking = False
+        if self._utterance_q is not None:
+            self._utterance_q.put_nowait(None)
+        self._utterance_q = None
+        self._event_ch.send_nowait(
+            stt.SpeechEvent(type=stt.SpeechEventType.END_OF_SPEECH)
+        )
+
+    async def _handshake(self, ws: websockets.ClientConnection) -> str:
+        created = json.loads(await ws.recv())
+        if created.get("type") != "session.created":
+            raise APIStatusError(
+                f"expected session.created, got {created}",
+                status_code=500,
+                body=created,
+            )
+        session_update: dict = {"type": "session.update", "model": self._opts.model}
+        if self._opts.target_streaming_delay_ms is not None:
+            session_update["target_streaming_delay_ms"] = (
+                self._opts.target_streaming_delay_ms
+            )
+        await ws.send(json.dumps(session_update))
+        return created.get("id", "")
+
+    def _auth_headers(self) -> dict[str, str]:
+        if self._opts.api_key:
+            return {"Authorization": f"Bearer {self._opts.api_key}"}
+        return {}
+
+    async def _utterance_pipeline(self) -> None:
+        # Owns the websocket lifecycle. On drop, reopens and resumes the
+        # in-flight utterance (if any) by replaying its already-sent chunks.
+        pending: _PendingUtterance | None = None
+        attempt = 0
+        while True:
+            try:
+                async with websockets.connect(
+                    self._opts.base_url,
+                    additional_headers=self._auth_headers(),
+                    open_timeout=self._conn_options.timeout,
+                ) as ws:
+                    request_id = await self._handshake(ws)
+                    attempt = 0
+                    while True:
+                        if pending is None:
+                            q = await self._utterance_chan.get()
+                            if q is None:
+                                return
+                            pending = _PendingUtterance(queue=q)
+                        await self._process_utterance(ws, pending, request_id)
+                        pending = None
+            except (websockets.WebSocketException, OSError, TimeoutError) as exc:
+                attempt += 1
+                if attempt > RECONNECT_MAX_ATTEMPTS:
+                    logger.exception(
+                        "vLLM realtime: giving up after %d reconnect attempts",
+                        RECONNECT_MAX_ATTEMPTS,
+                    )
+                    raise APIConnectionError() from exc
+                backoff = min(
+                    RECONNECT_BACKOFF_BASE_S * (2 ** (attempt - 1)),
+                    RECONNECT_BACKOFF_MAX_S,
+                )
+                if pending is None:
+                    logger.warning(
+                        "vLLM WS connection lost between utterances "
+                        "(attempt %d/%d): %s; retrying in %.1fs",
+                        attempt,
+                        RECONNECT_MAX_ATTEMPTS,
+                        exc,
+                        backoff,
+                    )
+                else:
+                    logger.warning(
+                        "vLLM WS dropped mid-utterance (%d chunks buffered, "
+                        "ended=%s, attempt %d/%d): %s; retrying in %.1fs",
+                        len(pending.sent_chunks),
+                        pending.ended,
+                        attempt,
+                        RECONNECT_MAX_ATTEMPTS,
+                        exc,
+                        backoff,
+                    )
+                await asyncio.sleep(backoff)
+
+    async def _process_utterance(
+        self,
+        ws: websockets.ClientConnection,
+        pending: _PendingUtterance,
+        request_id: str,
+    ) -> None:
+        # Start a fresh generation. Safe to send here: the previous utterance's
+        # transcription.done has already been received (we await it below), so
+        # the server-side generation_task is done and won't ignore this commit.
+        await ws.send(json.dumps({"type": "input_audio_buffer.commit"}))
+        send_t = asyncio.create_task(self._send_audio(ws, pending))
+        try:
+            await self._receive_one_transcription(ws, request_id)
+        finally:
+            if not send_t.done():
+                send_t.cancel()
+            try:
+                await send_t
+            except (asyncio.CancelledError, websockets.WebSocketException):
+                pass
+            except Exception:
+                logger.exception("send-audio task failed during finalize")
+
+    @staticmethod
+    async def _send_audio(
+        ws: websockets.ClientConnection, pending: _PendingUtterance
+    ) -> None:
+        # Replay anything already sent on a previous (now-dead) connection.
+        # sent_chunks is appended before send, so a chunk that failed to send
+        # last time is still present and gets retried here.
+        for chunk in pending.sent_chunks:
+            await ws.send(
+                json.dumps(
+                    {
+                        "type": "input_audio_buffer.append",
+                        "audio": base64.b64encode(chunk).decode("ascii"),
+                    }
+                )
+            )
+        if pending.ended:
+            await ws.send(
+                json.dumps({"type": "input_audio_buffer.commit", "final": True})
+            )
+            return
+        while True:
+            chunk = await pending.queue.get()
+            if chunk is None:
+                pending.ended = True
+                await ws.send(
+                    json.dumps({"type": "input_audio_buffer.commit", "final": True})
+                )
+                return
+            pending.sent_chunks.append(chunk)
+            await ws.send(
+                json.dumps(
+                    {
+                        "type": "input_audio_buffer.append",
+                        "audio": base64.b64encode(chunk).decode("ascii"),
+                    }
+                )
+            )
+
+    async def _receive_one_transcription(
+        self, ws: websockets.ClientConnection, request_id: str
+    ) -> None:
+        # Use recv() rather than `async for`: the latter swallows
+        # ConnectionClosed on close-mid-iteration, which would let a dropped
+        # WS look like a clean "no transcription" return.
+        current_text = ""
+        while True:
+            raw = await ws.recv()
+            data = json.loads(raw)
+            event_type = data.get("type")
+
+            if event_type == "transcription.delta":
+                delta = data.get("delta", "")
+                if not delta:
+                    continue
+                current_text += delta
+                self._event_ch.send_nowait(
+                    stt.SpeechEvent(
+                        type=stt.SpeechEventType.INTERIM_TRANSCRIPT,
+                        request_id=request_id,
+                        alternatives=[stt.SpeechData(text=current_text, language="")],
+                    )
+                )
+            elif event_type == "transcription.done":
+                final_text = data.get("text") or current_text
+                self._event_ch.send_nowait(
+                    stt.SpeechEvent(
+                        type=stt.SpeechEventType.FINAL_TRANSCRIPT,
+                        request_id=request_id,
+                        alternatives=[stt.SpeechData(text=final_text, language="")],
+                    )
+                )
+                usage = data.get("usage") or {}
+                self._event_ch.send_nowait(
+                    stt.SpeechEvent(
+                        type=stt.SpeechEventType.RECOGNITION_USAGE,
+                        request_id=request_id,
+                        recognition_usage=stt.RecognitionUsage(
+                            audio_duration=float(
+                                usage.get("audio_seconds")
+                                or usage.get("prompt_audio_seconds")
+                                or 0
+                            ),
+                            input_tokens=int(usage.get("prompt_tokens") or 0),
+                            output_tokens=int(usage.get("completion_tokens") or 0),
+                        ),
+                    )
+                )
+                return
+            elif event_type == "error":
+                err = data.get("error")
+                raise APIStatusError(str(err), status_code=500, body=data)

src/backend/core/api/feature_flag.py

@@ -11,7 +11,6 @@ class FeatureFlag:
 
     FLAGS = {
         "recording": "RECORDING_ENABLE",
-        "storage_event": "RECORDING_STORAGE_EVENT_ENABLE",
         "subtitle": "ROOM_SUBTITLE_ENABLED",
         "file_upload": "FILE_UPLOAD_ENABLED",
         "addons": "ADDONS_ENABLED",

src/backend/core/api/viewsets.py

@@ -45,25 +45,11 @@ from core.api import throttling
 from core.api.filters import ListFileFilter
 from core.enums import MEDIA_STORAGE_URL_PATTERN
 from core.recording.enums import FileExtension
-from core.recording.event.authentication import (
-    RecordingProcessWebhookAuthentication,
-    StorageEventAuthentication,
-)
-from core.recording.event.exceptions import (
-    InvalidBucketError,
-    InvalidFilepathError,
-    InvalidFileTypeError,
-    ParsingEventDataError,
-)
-from core.recording.event.parsers import get_parser
+from core.recording.event.authentication import RecordingProcessWebhookAuthentication
 from core.recording.services.metadata_collector import (
     MetadataCollectorException,
     MetadataCollectorService,
 )
-from core.recording.services.recording_events import (
-    RecordingEventsService,
-    RecordingNotSavableError,
-)
 from core.recording.worker.exceptions import (
     RecordingStartError,
     RecordingStopError,
@@ -1034,56 +1020,6 @@ class RecordingViewSet(
             .filter(Q(accesses__user=user) | Q(accesses__team__in=user.get_teams()))
         )
 
-    @decorators.action(
-        detail=False,
-        methods=["post"],
-        url_path="storage-hook",
-        authentication_classes=[StorageEventAuthentication],
-    )
-    @FeatureFlag.require("storage_event")
-    def on_storage_event_received(self, request, pk=None):  # pylint: disable=unused-argument
-        """Handle incoming storage hook events for recordings."""
-
-        parser = get_parser()
-
-        try:
-            recording_id = parser.get_recording_id(request.data)
-
-        except ParsingEventDataError as e:
-            raise drf_exceptions.PermissionDenied("Invalid request data.") from e
-
-        except InvalidBucketError as e:
-            raise drf_exceptions.PermissionDenied("Invalid bucket specified.") from e
-
-        except InvalidFilepathError:
-            return drf_response.Response(
-                {"message": "Notification ignored."},
-            )
-
-        except InvalidFileTypeError:
-            return drf_response.Response(
-                {"message": "Notification ignored."},
-            )
-
-        try:
-            recording = models.Recording.objects.get(id=recording_id)
-        except models.Recording.DoesNotExist as e:
-            raise drf_exceptions.NotFound("No recording found for this event.") from e
-
-        # Save recording
-        recording_events_service = RecordingEventsService()
-        try:
-            recording_events_service.handle_complete(recording)
-        except RecordingNotSavableError:
-            raise drf_exceptions.PermissionDenied(
-                f"Recording with ID {recording_id} cannot be saved because it is either,"
-                " in an error state or has already been saved."
-            ) from None
-
-        return drf_response.Response(
-            {"message": "Event processed."},
-        )
-
     @decorators.action(
         detail=False,
         methods=["post"],

src/backend/core/recording/event/authentication.py

@@ -1,4 +1,4 @@
-"""Authentication class for storage event token validation."""
+"""Authentication classes for server-to-server webhook token validation."""
 
 import logging
 import secrets
@@ -12,9 +12,9 @@ logger = logging.getLogger(__name__)
 
 
 class MachineUser:
-    """Represent a non-interactive system user for automated storage operations."""
+    """Represent a non-interactive system user for automated operations."""
 
-    def __init__(self, username: str = "storage_event_user") -> None:
+    def __init__(self, username: str = "machine_user") -> None:
         self.pk = None
         self.username = username
         self.is_active = True
@@ -41,24 +41,17 @@ class HeaderBasedAuthentication(BaseAuthentication):
     TOKEN_TYPE = "Bearer"  # noqa S105
     REALM = ""
 
-    IS_ENFORCED_SETTINGS_KEY = None
     EXPECTED_TOKEN_SETTINGS_KEY = None
 
     def authenticate(self, request):
         """Validate the Bearer token from the Authorization header."""
 
-        if self.IS_ENFORCED_SETTINGS_KEY is not None:
-            if not getattr(settings, self.IS_ENFORCED_SETTINGS_KEY):
-                return MachineUser(), None
-
         if (
             self.EXPECTED_TOKEN_SETTINGS_KEY is None
             or (required_token := getattr(settings, self.EXPECTED_TOKEN_SETTINGS_KEY))
             is None
         ):
-            raise AuthenticationFailed(
-                "Authentication is enabled but token is not configured."
-            )
+            raise AuthenticationFailed("Authentication token is not configured.")
 
         auth_header = request.headers.get(self.AUTH_HEADER)
         if not auth_header:
@@ -88,18 +81,6 @@ class HeaderBasedAuthentication(BaseAuthentication):
         return f"{self.TOKEN_TYPE} realm='{self.REALM}'"
 
 
-class StorageEventAuthentication(HeaderBasedAuthentication):
-    """Authenticate requests using a Bearer token for storage event integration.
-    This class validates Bearer tokens for storage events that don't map to database users.
-    It's designed for S3-compatible storage integrations and similar use cases.
-    Events are submitted when a webhook is configured on some bucket's events.
-    """
-
-    REALM = "Storage event API"
-    IS_ENFORCED_SETTINGS_KEY = "RECORDING_ENABLE_STORAGE_EVENT_AUTH"
-    EXPECTED_TOKEN_SETTINGS_KEY = "RECORDING_STORAGE_EVENT_TOKEN"  # noqa S105
-
-
 class RecordingProcessWebhookAuthentication(HeaderBasedAuthentication):
     """
     Custom authentication class for recording process webhook requests.

src/backend/core/recording/event/exceptions.py

@@ -1,17 +0,0 @@
-"""Storage parsers specific exceptions."""
-
-
-class ParsingEventDataError(Exception):
-    """Raised when the request data is malformed, incomplete, or missing."""
-
-
-class InvalidBucketError(Exception):
-    """Raised when the bucket name in the request does not match the expected one."""
-
-
-class InvalidFileTypeError(Exception):
-    """Raised when the file type in the request is not supported."""
-
-
-class InvalidFilepathError(Exception):
-    """Raised when the filepath in the request is invalid."""

src/backend/core/recording/event/parsers.py

@@ -1,178 +0,0 @@
-"""Meet storage event parser classes."""
-
-import logging
-import mimetypes
-import re
-from dataclasses import dataclass
-from functools import lru_cache
-from typing import Any, Dict, Optional, Protocol
-from urllib.parse import quote
-
-from django.conf import settings
-from django.utils.module_loading import import_string
-
-from core.enums import FILE_EXT_REGEX, UUID_REGEX
-
-from .exceptions import (
-    InvalidBucketError,
-    InvalidFilepathError,
-    InvalidFileTypeError,
-    ParsingEventDataError,
-)
-
-# Additional MIME type mapping
-mimetypes.add_type("audio/ogg", ".ogg")
-
-logger = logging.getLogger(__name__)
-
-
-@dataclass
-class StorageEvent:
-    """Represents a storage event with relevant metadata.
-    Attributes:
-        filepath: Identifier for the affected recording
-        filetype: Type of storage event
-        bucket_name: When the event occurred
-        metadata: Additional event data
-    """
-
-    filepath: str
-    filetype: str
-    bucket_name: str
-    metadata: Optional[Dict[str, Any]]
-
-    def __post_init__(self):
-        if self.filepath is None:
-            raise TypeError("filepath cannot be None")
-        if self.filetype is None:
-            raise TypeError("filetype cannot be None")
-        if self.bucket_name is None:
-            raise TypeError("bucket_name cannot be None")
-
-
-class EventParser(Protocol):
-    """Interface for parsing storage events."""
-
-    def __init__(self, bucket_name, allowed_filetypes=None):
-        """Initialize parser with bucket name and optional allowed filetypes."""
-
-    def parse(self, data: Dict) -> StorageEvent:
-        """Extract storage event data from raw dictionary input."""
-
-    def validate(self, data: StorageEvent) -> str:
-        """Verify storage event data meets all requirements."""
-
-    def get_recording_id(self, data: Dict) -> str:
-        """Extract recording ID from event dictionary."""
-
-
-@lru_cache(maxsize=1)
-def get_parser() -> EventParser:
-    """Return cached instance of configured event parser.
-    Uses function memoization instead of a factory class since the only
-    varying parameter is the parser class from settings. A factory class
-    would add unnecessary complexity when a cached function provides the
-    same singleton behavior with simpler code.
-    """
-
-    event_parser_cls = import_string(settings.RECORDING_EVENT_PARSER_CLASS)
-    return event_parser_cls(bucket_name=settings.AWS_STORAGE_BUCKET_NAME)
-
-
-class BaseS3Parser:
-    """Base class for handling parsing and validation of S3-compatible storage events."""
-
-    def __init__(self, bucket_name: str, allowed_filetypes=None):
-        """Initialize parser with target bucket name and accepted filetypes."""
-
-        if not bucket_name:
-            raise ValueError("Bucket name cannot be None or empty")
-
-        self._bucket_name = bucket_name
-        self._allowed_filetypes = allowed_filetypes or {"audio/ogg", "video/mp4"}
-
-        # pylint: disable=line-too-long
-        self._filepath_regex = re.compile(
-            rf"(?P<url_encoded_folder_path>(?:[^%]+%2F)+)?{settings.RECORDING_OUTPUT_FOLDER}%2F(?P<recording_id>{UUID_REGEX})\.(?P<extension>{FILE_EXT_REGEX})"
-        )
-
-    def validate(self, event_data: StorageEvent) -> str:
-        """Verify StorageEvent matches bucket, filetype and filepath requirements."""
-
-        if event_data.bucket_name != self._bucket_name:
-            raise InvalidBucketError(
-                f"Invalid bucket: expected {self._bucket_name}, got {event_data.bucket_name}"
-            )
-
-        if event_data.filetype not in self._allowed_filetypes:
-            raise InvalidFileTypeError(
-                f"Invalid file type, expected {self._allowed_filetypes},"
-                f"got '{event_data.filetype}'"
-            )
-
-        match = self._filepath_regex.match(event_data.filepath)
-        if not match:
-            raise InvalidFilepathError(
-                f"Invalid filepath structure: {event_data.filepath}"
-            )
-
-        recording_id = match.group("recording_id")
-        return recording_id
-
-    def get_recording_id(self, data):
-        """Extract recording ID from S3 event through parsing and validation."""
-
-        event_data = self.parse(data)
-        return self.validate(event_data)
-
-    def parse(self, data: Dict) -> StorageEvent:
-        """To be implemented by subclasses."""
-        raise NotImplementedError("Subclasses must implement parse()")
-
-
-class MinioParser(BaseS3Parser):
-    """Minio specific event parsing."""
-
-    def parse(self, data: Dict) -> StorageEvent:
-        if not data:
-            raise ParsingEventDataError("Received empty data.")
-        try:
-            record = data["Records"][0]
-            s3 = record["s3"]
-            return StorageEvent(
-                filepath=s3["object"]["key"],
-                filetype=s3["object"]["contentType"],  # Minio-specific field
-                bucket_name=s3["bucket"]["name"],
-                metadata=None,
-            )
-        except (KeyError, IndexError) as e:
-            raise ParsingEventDataError(f"Malformed Minio event: {e}") from e
-        except TypeError as e:
-            raise ParsingEventDataError(f"Missing essential data fields: {e}") from e
-
-
-class S3Parser(BaseS3Parser):
-    """AWS S3 specific event parsing."""
-
-    def parse(self, data: Dict) -> StorageEvent:
-        if not data:
-            raise ParsingEventDataError("Received empty data.")
-        try:
-            # AWS S3 structure can slightly differ from Minio implementation
-            record = data["Records"][0]
-            s3 = record["s3"]
-            filepath = s3["object"]["key"]
-            if not filepath:
-                raise ParsingEventDataError("Missing object key name")
-            filetype, _ = mimetypes.guess_type(filepath)
-            # Normalize raw S3-compatible object keys without re-encoding
-            # already encoded AWS S3 notification keys.
-            filepath = quote(filepath, safe="%+")
-            return StorageEvent(
-                filepath=filepath,
-                filetype=filetype,
-                bucket_name=s3["bucket"]["name"],
-                metadata=None,
-            )
-        except (KeyError, IndexError) as e:
-            raise ParsingEventDataError(f"Malformed S3 event: {e}") from e

src/backend/core/services/livekit_events.py

@@ -220,10 +220,8 @@ class LiveKitEventsService:
                     f"Failed to process limit reached event for recording {recording}"
                 ) from e
 
-        # Fallback for completion when no MinIO/S3 webhooks are configured
-        if (
-            not settings.RECORDING_STORAGE_EVENT_ENABLE
-        ) and data.egress_info.status in [
+        # Finalize the recording, the egress has uploaded the file to the storage
+        if data.egress_info.status in [
             api.EgressStatus.EGRESS_COMPLETE,
             api.EgressStatus.EGRESS_LIMIT_REACHED,
         ]:

src/backend/core/tests/recording/event/test_authentication.py

@@ -11,135 +11,98 @@ from rest_framework.exceptions import AuthenticationFailed
 
 from core.recording.event.authentication import (
     MachineUser,
-    StorageEventAuthentication,
+    RecordingProcessWebhookAuthentication,
 )
 
 
 def test_successful_authentication(settings):
     """Test successful authentication with valid token."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "Bearer valid-test-token"}
 
-    user, token = StorageEventAuthentication().authenticate(request)
+    user, token = RecordingProcessWebhookAuthentication().authenticate(request)
     assert token == "valid-test-token"
     assert isinstance(user, MachineUser)
 
 
-def test_disabled_authentication_with_header(settings):
-    """Authentication should pass when no auth is configured, and header is present."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = None
-    settings.RECORDING_ENABLE_STORAGE_EVENT_AUTH = False
-
-    request = RequestFactory().get("/")
-    request.headers = {"Authorization": "Bearer some-token"}
-
-    user, token = StorageEventAuthentication().authenticate(request)
-    assert token is None
-    assert isinstance(user, MachineUser)
-
-
-def test_disabled_authentication_without_header(settings):
-    """Authentication should pass when no auth is configured, and no header is present."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = None
-    settings.RECORDING_ENABLE_STORAGE_EVENT_AUTH = False
-
-    request = RequestFactory().get("/")
-
-    user, token = StorageEventAuthentication().authenticate(request)
-    assert token is None
-    assert isinstance(user, MachineUser)
-
-
-def test_authentication_when_disabled(settings):
-    """Authentication should pass when disabled, regardless of token configuration."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "some-token"
-    settings.RECORDING_ENABLE_STORAGE_EVENT_AUTH = False
-
-    request = RequestFactory().get("/")
-
-    user, token = StorageEventAuthentication().authenticate(request)
-    assert token is None
-    assert isinstance(user, MachineUser)
-
-
 def test_authentication_fails_when_token_not_configured(settings):
-    """Authentication should fail when authentication is enabled but no token is configured."""
+    """Authentication should fail when no token is configured."""
 
-    # By default RECORDING_ENABLE_STORAGE_EVENT_AUTH should be True
-    settings.RECORDING_STORAGE_EVENT_TOKEN = None
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = None
 
     request = RequestFactory().get("/")
 
     with pytest.raises(
         AuthenticationFailed,
-        match="Authentication is enabled but token is not configured",
+        match="Authentication token is not configured",
     ):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_missing_auth_header(settings):
     """Test failure when Authorization header is missing."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {}
 
     with pytest.raises(AuthenticationFailed, match="Authorization header is required"):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_invalid_auth_header_format(settings):
     """Test failure when Authorization header has invalid format."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "InvalidFormat"}
 
     with pytest.raises(AuthenticationFailed, match="Invalid authorization header"):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_invalid_token_type(settings):
     """Test failure when token type is not Bearer."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "Basic some-token"}
 
     with pytest.raises(AuthenticationFailed, match="Invalid authorization header"):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_invalid_token(settings):
     """Test failure when token is invalid."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "Bearer wrong-token"}
 
     with pytest.raises(AuthenticationFailed, match="Invalid token"):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_malformed_auth_header(settings):
     """Test failure when Authorization header is malformed."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "valid-test-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "Bearer"}  # Missing token part
 
     with pytest.raises(AuthenticationFailed, match="Invalid authorization header"):
-        StorageEventAuthentication().authenticate(request)
+        RecordingProcessWebhookAuthentication().authenticate(request)
 
 
 def test_authenticate_header():
     """Test the WWW-Authenticate header value."""
     request = RequestFactory().get("/")
-    header = StorageEventAuthentication().authenticate_header(request)
-    assert header == "Bearer realm='Storage event API'"
+    header = RecordingProcessWebhookAuthentication().authenticate_header(request)
+    assert header == "Bearer realm='External process webhook API'"
 
 
 def test_multiple_spaces_in_auth_header(settings):
-    """Test success when Authorization header contains multiple spaces."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "valid-test-token"
+    """Extra spaces between the scheme and the token should be tolerated."""
+    settings.SUMMARY_SERVICE_WEBHOOK_API_TOKEN = "extra-spaces-token"
     request = RequestFactory().get("/")
     request.headers = {"Authorization": "Bearer   extra-spaces-token"}
 
-    header = StorageEventAuthentication().authenticate_header(request)
-    assert header == "Bearer realm='Storage event API'"
+    user, token = RecordingProcessWebhookAuthentication().authenticate(request)
+    assert token == "extra-spaces-token"
+    assert isinstance(user, MachineUser)

src/backend/core/tests/recording/event/test_parsers.py

@@ -1,512 +0,0 @@
-"""
-Test event parsers.
-"""
-
-# pylint: disable=protected-access,redefined-outer-name,unused-argument
-
-from unittest import mock
-
-from django.conf import settings
-
-import pytest
-
-from core.recording.event.exceptions import (
-    InvalidBucketError,
-    InvalidFilepathError,
-    InvalidFileTypeError,
-    ParsingEventDataError,
-)
-from core.recording.event.parsers import (
-    MinioParser,
-    S3Parser,
-    StorageEvent,
-    get_parser,
-)
-
-# MinioParser
-
-
-@pytest.fixture
-def valid_minio_event():
-    """Mock a valid Minio event."""
-    return {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-                        "contentType": "audio/ogg",
-                    },
-                }
-            }
-        ]
-    }
-
-
-@pytest.fixture
-def minio_parser():
-    """Mock a Minio parser."""
-    return MinioParser(bucket_name="test-bucket")
-
-
-def test_minio_parse_valid_event(minio_parser, valid_minio_event):
-    """Test parsing a valid Minio event."""
-    event = minio_parser.parse(valid_minio_event)
-    assert isinstance(event, StorageEvent)
-    assert event.filepath == "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg"
-    assert event.filetype == "audio/ogg"
-    assert event.bucket_name == "test-bucket"
-    assert event.metadata is None
-
-
-def test_minio_parse_with_video_type(minio_parser):
-    """Test parsing event with video file type."""
-    video_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "46d1a121-2426-484d-8fb3-09b5d886f7a8.mp4",
-                        "contentType": "video/mp4",
-                    },
-                }
-            }
-        ]
-    }
-    event = minio_parser.parse(video_event)
-    assert event.filetype == "video/mp4"
-    assert event.filepath.endswith(".mp4")
-
-
-def test_minio_parse_empty_data(minio_parser):
-    """Test parsing empty event data raises error."""
-    with pytest.raises(ParsingEventDataError, match="Received empty data."):
-        minio_parser.parse({})
-
-
-def test_minio_parse_missing_keys(minio_parser):
-    """Test parsing event with missing key."""
-
-    invalid_minio_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": None},
-                    # Missing 'object' key
-                }
-            }
-        ]
-    }
-
-    with pytest.raises(ParsingEventDataError, match="Malformed Minio event:"):
-        minio_parser.parse(invalid_minio_event)
-
-
-def test_minio_parse_none_key(minio_parser):
-    """Test parsing event with None field."""
-
-    invalid_minio_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "recording%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-                        "contentType": None,  # 'contentType' should not be None
-                    },
-                }
-            }
-        ]
-    }
-
-    with pytest.raises(ParsingEventDataError, match="Missing essential data fields"):
-        minio_parser.parse(invalid_minio_event)
-
-
-def test_minio_validate_invalid_bucket(minio_parser):
-    """Test validation with wrong bucket name."""
-    event = StorageEvent(
-        filepath="recording%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-        filetype="audio/ogg",
-        bucket_name="wrong-bucket",
-        metadata=None,
-    )
-    with pytest.raises(InvalidBucketError):
-        minio_parser.validate(event)
-
-
-def test_minio_validate_invalid_filetype(minio_parser):
-    """Test validation with unsupported file type."""
-    event = StorageEvent(
-        filepath="recording%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.txt",
-        filetype="text/plain",  # Not included in the default allowed filetypes
-        bucket_name="test-bucket",
-        metadata=None,
-    )
-    with pytest.raises(InvalidFileTypeError):
-        minio_parser.validate(event)
-
-
-@pytest.mark.parametrize(
-    "invalid_filepath",
-    [
-        "invalid_filepath",  # totally invalid string
-        "recordings/46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-        "recordings/46d1a121-2426-484d-8fb3-09b5d886f7a8",  # missing extension
-        "46d1a121-2426-484d-8fb3-09b5d886f7a8",  # missing url_encoded_folder_path and extension
-        "",  # empty string
-        "46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",  # no folder at all
-        "uploads%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",  # wrong folder name
-        "folder%2Fuploads%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",  # nested but no recordings/
-    ],
-)
-def test_minio_validate_invalid_filepath(invalid_filepath, minio_parser):
-    """Test validation with malformed filepath."""
-    event = StorageEvent(
-        filepath=invalid_filepath,
-        filetype="audio/ogg",
-        bucket_name="test-bucket",
-        metadata=None,
-    )
-    with pytest.raises(InvalidFilepathError):
-        minio_parser.validate(event)
-
-
-def test_minio_validate_valid_event(minio_parser):
-    """Test validation with valid event data."""
-    event = StorageEvent(
-        filepath="recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-        filetype="audio/ogg",
-        bucket_name="test-bucket",
-        metadata=None,
-    )
-    recording_id = minio_parser.validate(event)
-    assert recording_id == "46d1a121-2426-484d-8fb3-09b5d886f7a8"
-
-
-def test_minio_get_recording_id_success(minio_parser, valid_minio_event):
-    """Test successful extraction of recording ID."""
-    recording_id = minio_parser.get_recording_id(valid_minio_event)
-    assert recording_id == "46d1a121-2426-484d-8fb3-09b5d886f7a8"
-
-
-def test_minio_validate_filepath_with_folder(minio_parser):
-    """Test validation of filepath with folder structure."""
-    event = StorageEvent(
-        filepath="parent_folder%2Frecordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-        filetype="audio/ogg",
-        bucket_name="test-bucket",
-        metadata=None,
-    )
-    recording_id = minio_parser.validate(event)
-    assert recording_id == "46d1a121-2426-484d-8fb3-09b5d886f7a8"
-
-
-def test_minio_empty_allowed_filetypes():
-    """Test MinioParser with empty allowed_filetypes."""
-    empty_types = set()
-    parser = MinioParser(bucket_name="test-bucket", allowed_filetypes=empty_types)
-    assert parser._allowed_filetypes == {"audio/ogg", "video/mp4"}
-
-
-def test_minio_custom_allowed_filetypes():
-    """Test MinioParser with empty allowed_filetypes."""
-    custom_types = {"audio/mp3", "video/mov"}
-    parser = MinioParser(bucket_name="test-bucket", allowed_filetypes=custom_types)
-    assert parser._allowed_filetypes == {"audio/mp3", "video/mov"}
-
-
-def test_minio_validate_custom_filetypes():
-    """Test validation of filepath with folder structure."""
-
-    parser = MinioParser(bucket_name="test-bucket", allowed_filetypes={"audio/mp3"})
-
-    event = StorageEvent(
-        filepath="parent_folder%2Frecordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-        filetype="audio/mp3",
-        bucket_name="test-bucket",
-        metadata=None,
-    )
-    parser.validate(event)
-
-
-def test_minio_constructor_none_bucket():
-    """Test MinioParser constructor with None bucket name."""
-    with pytest.raises(ValueError, match="Bucket name cannot be None or empty"):
-        MinioParser(bucket_name=None)
-
-
-def test_minio_constructor_empty_bucket():
-    """Test MinioParser constructor with empty bucket name."""
-    with pytest.raises(ValueError, match="Bucket name cannot be None or empty"):
-        MinioParser(bucket_name="")
-
-
-# S3Parser
-
-
-@pytest.fixture
-def valid_s3_event():
-    """Mock a valid S3 event."""
-    return {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg",
-                    },
-                }
-            }
-        ]
-    }
-
-
-@pytest.fixture
-def s3_parser():
-    """Mock an S3 parser."""
-    return S3Parser(bucket_name="test-bucket")
-
-
-def test_s3_parse_valid_event(s3_parser, valid_s3_event):
-    """Test parsing a valid S3 event."""
-    event = s3_parser.parse(valid_s3_event)
-    assert isinstance(event, StorageEvent)
-    assert event.filepath == "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.ogg"
-    assert event.filetype == "audio/ogg"
-    assert event.bucket_name == "test-bucket"
-    assert event.metadata is None
-
-
-def test_s3_parse_empty_data(s3_parser):
-    """Test parsing empty S3 event data raises error."""
-    with pytest.raises(ParsingEventDataError, match="Received empty data."):
-        s3_parser.parse({})
-
-
-def test_s3_parse_missing_keys(s3_parser):
-    """Test parsing S3 event with missing key."""
-    invalid_s3_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    # Missing 'object' key
-                }
-            }
-        ]
-    }
-
-    with pytest.raises(ParsingEventDataError, match="Malformed S3 event:"):
-        s3_parser.parse(invalid_s3_event)
-
-
-def test_s3_parse_none_key(s3_parser):
-    """Test parsing S3 event with None field."""
-    invalid_s3_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": None,
-                    },
-                }
-            }
-        ]
-    }
-
-    with pytest.raises(ParsingEventDataError, match="Missing object key name"):
-        s3_parser.parse(invalid_s3_event)
-
-
-def test_s3_parse_with_video_type(s3_parser):
-    """Test parsing S3 event with mp4 file extension."""
-    video_event = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.mp4",
-                    },
-                }
-            }
-        ]
-    }
-    event = s3_parser.parse(video_event)
-    assert event.filetype == "video/mp4"
-    assert event.filepath.endswith(".mp4")
-
-
-def test_s3_parse_unrecognized_extension(s3_parser):
-    """Test parsing S3 event with unrecognized file extension."""
-    event_with_unknown_ext = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "test-bucket"},
-                    "object": {
-                        "key": "recordings%2F46d1a121-2426-484d-8fb3-09b5d886f7a8.zzunknown999",
-                    },
-                }
-            }
-        ]
-    }
-
-    with pytest.raises(TypeError, match="filetype cannot be None"):
-        s3_parser.parse(event_with_unknown_ext)
-
-
-def test_s3_parser_keeps_encoded_filepath_compatible(settings):
-    """Test S3 parser keeps already encoded object keys compatible."""
-    settings.RECORDING_OUTPUT_FOLDER = "recordings"
-
-    recording_id = "80ae9fe5-639a-438b-b86e-9e3dd2d55f4d"
-    parser = S3Parser(bucket_name="recordings-bucket")
-
-    data = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "recordings-bucket"},
-                    "object": {
-                        "key": f"recordings%2F{recording_id}.mp4",
-                    },
-                }
-            }
-        ]
-    }
-
-    assert parser.get_recording_id(data) == recording_id
-
-
-def test_s3_parser_accepts_unencoded_filepath(settings):
-    """Test S3 parser accepts raw object keys with slash separators."""
-    settings.RECORDING_OUTPUT_FOLDER = "recordings"
-
-    recording_id = "80ae9fe5-639a-438b-b86e-9e3dd2d55f4d"
-    parser = S3Parser(bucket_name="recordings-bucket")
-
-    data = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "recordings-bucket"},
-                    "object": {
-                        "key": f"recordings/{recording_id}.mp4",
-                    },
-                }
-            }
-        ]
-    }
-
-    assert parser.get_recording_id(data) == recording_id
-
-
-def test_s3_parser_preserves_plus_signs_in_encoded_filepath(settings):
-    """Test S3 parser preserves plus signs in already encoded object keys."""
-    settings.RECORDING_OUTPUT_FOLDER = "recordings"
-
-    recording_id = "80ae9fe5-639a-438b-b86e-9e3dd2d55f4d"
-    parser = S3Parser(bucket_name="recordings-bucket")
-
-    data = {
-        "Records": [
-            {
-                "s3": {
-                    "bucket": {"name": "recordings-bucket"},
-                    "object": {
-                        "key": f"folder+name%2Frecordings%2F{recording_id}.mp4",
-                    },
-                }
-            }
-        ]
-    }
-
-    assert parser.get_recording_id(data) == recording_id
-
-
-def test_s3_get_recording_id_success(s3_parser, valid_s3_event):
-    """Test successful extraction of recording ID from S3 event."""
-    recording_id = s3_parser.get_recording_id(valid_s3_event)
-    assert recording_id == "46d1a121-2426-484d-8fb3-09b5d886f7a8"
-
-
-# get_parser
-
-
-@pytest.fixture
-def clear_lru_cache():
-    """Fixture to clear the LRU cache between tests."""
-    get_parser.cache_clear()
-    yield
-    get_parser.cache_clear()
-
-
-def test_returns_correct_instance(clear_lru_cache):
-    """Test if get_parser returns the correct parser instance."""
-    settings.AWS_STORAGE_BUCKET_NAME = "test-bucket"
-    parser = get_parser()
-    assert isinstance(parser, MinioParser)
-    assert parser._bucket_name == "test-bucket"
-
-
-def test_caching_behavior(clear_lru_cache):
-    """Test if the function properly caches the parser instance."""
-    settings.AWS_STORAGE_BUCKET_NAME = "test-bucket"
-    parser1 = get_parser()
-    parser2 = get_parser()
-    assert parser1 is parser2  # Check object identity
-
-
-def test_different_settings_new_instance():
-    """Test if changing settings creates a new instance."""
-    settings.AWS_STORAGE_BUCKET_NAME = "different-bucket"
-    parser = get_parser()
-    assert parser._bucket_name == "different-bucket"
-
-
-def test_import_error_handling(clear_lru_cache):
-    """Test handling of import errors for invalid parser class."""
-    settings.RECORDING_EVENT_PARSER_CLASS = "invalid.parser.path"
-    with pytest.raises(ImportError):
-        get_parser()
-
-
-@mock.patch("core.recording.event.parsers.import_string")
-def test_parser_instantiation_called_once(mock_import_string, clear_lru_cache):
-    """Test that parser class is instantiated only once due to caching."""
-    mock_parser_cls = type(
-        "MockParser",
-        (),
-        {
-            "__init__": lambda self, bucket_name: setattr(
-                self, "_bucket_name", bucket_name
-            )
-        },
-    )
-    mock_import_string.return_value = mock_parser_cls
-
-    # First call
-    parser1 = get_parser()
-    # Second call
-    parser2 = get_parser()
-
-    # Verify import_string was called only once
-    mock_import_string.assert_called_once_with(settings.RECORDING_EVENT_PARSER_CLASS)
-    assert parser1 is parser2
-
-
-def test_cache_clear_behavior(clear_lru_cache, settings):
-    """Test that cache clearing creates new instance."""
-
-    settings.RECORDING_EVENT_PARSER_CLASS = "core.recording.event.parsers.MinioParser"
-
-    parser1 = get_parser()
-    get_parser.cache_clear()
-    parser2 = get_parser()
-
-    assert parser1 is not parser2  # Should be different instances after cache clear

src/backend/core/tests/recording/service/test_recording_events.py

@@ -12,6 +12,7 @@ from core.factories import RecordingFactory
 from core.recording.services.recording_events import (
     RecordingEventsError,
     RecordingEventsService,
+    RecordingNotSavableError,
 )
 from core.utils import NotificationError
 
@@ -70,3 +71,56 @@ def test_handle_limit_reached_error(mock_notify, mode, notification_type, servic
     mock_notify.assert_called_once_with(
         room_name=str(recording.room.id), notification_data={"type": notification_type}
     )
+
+
+@pytest.mark.parametrize("status", ["active", "stopped"])
+@pytest.mark.parametrize(
+    ("notify_return_value", "expected_status"),
+    ((True, "notification_succeeded"), (False, "saved")),
+)
+@mock.patch(
+    "core.recording.services.recording_events.notification_service."
+    "notify_external_services"
+)
+def test_handle_complete_saves_recording(  # pylint: disable=too-many-arguments, too-many-positional-arguments
+    mock_notify_external_services,
+    notify_return_value,
+    expected_status,
+    status,
+    service,
+):
+    """Test handle_complete notifies external services and saves a savable recording."""
+
+    mock_notify_external_services.return_value = notify_return_value
+
+    recording = RecordingFactory(status=status)
+    service.handle_complete(recording)
+
+    mock_notify_external_services.assert_called_once_with(recording)
+
+    recording.refresh_from_db()
+    assert recording.status == expected_status
+
+
+@pytest.mark.parametrize(
+    "status",
+    ["initiated", "saved", "notification_succeeded", "aborted", "failed_to_start"],
+)
+@mock.patch(
+    "core.recording.services.recording_events.notification_service."
+    "notify_external_services"
+)
+def test_handle_complete_non_savable_recording(
+    mock_notify_external_services, status, service
+):
+    """Test handle_complete refuses recordings that are already saved or in error."""
+
+    recording = RecordingFactory(status=status)
+
+    with pytest.raises(RecordingNotSavableError):
+        service.handle_complete(recording)
+
+    mock_notify_external_services.assert_not_called()
+
+    recording.refresh_from_db()
+    assert recording.status == status

src/backend/core/tests/recording/test_api_recordings_storage_hook.py

@@ -1,267 +0,0 @@
-"""
-Test recordings API endpoints in the Meet core app: save recording.
-"""
-
-# pylint: disable=redefined-outer-name,unused-argument
-
-import uuid
-from unittest import mock
-
-import pytest
-from rest_framework.test import APIClient
-
-from ...factories import RecordingFactory
-from ...models import Recording, RecordingStatusChoices
-from ...recording.event.exceptions import (
-    InvalidBucketError,
-    InvalidFilepathError,
-    InvalidFileTypeError,
-    ParsingEventDataError,
-)
-
-pytestmark = pytest.mark.django_db
-
-
-@pytest.fixture
-def recording_settings(settings):
-    """Configure recording-related and storage event Django settings."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "testAuthToken"
-    settings.RECORDING_STORAGE_EVENT_ENABLE = True
-    return settings
-
-
-@pytest.fixture
-def mock_get_parser():
-    """Mock 'get_parser' factory function."""
-    with mock.patch("core.api.viewsets.get_parser") as mock_parser:
-        yield mock_parser
-
-
-def test_save_recording_anonymous(settings, client):
-    """Anonymous users should not be allowed to save room recordings."""
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "testAuthToken"
-
-    RecordingFactory(status="active")
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-    )
-
-    assert response.status_code == 401
-    assert Recording.objects.count() == 1
-
-
-def test_save_recording_wrong_bearer(settings, client):
-    """Requests with incorrect bearer token should be rejected when auth is required."""
-
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "testAuthToken"
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer wrongAuthToken",
-    )
-
-    assert response.status_code == 401
-
-
-def test_save_recording_permission_needed(settings, client):
-    """Recordings should not be saved when feature is disabled."""
-
-    settings.RECORDING_STORAGE_EVENT_TOKEN = "testAuthToken"
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 404
-    assert response.json() == {"detail": "Not found."}
-
-
-def test_save_recording_parsing_error(recording_settings, mock_get_parser, client):
-    """Test handling of parsing errors in recording event data."""
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.side_effect = ParsingEventDataError("Error message")
-    mock_get_parser.return_value = mock_parser
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 403
-    assert response.json() == {"detail": "Invalid request data."}
-
-
-def test_save_recording_bucket_error(recording_settings, mock_get_parser, client):
-    """Test handling of invalid storage bucket errors in recording event data."""
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.side_effect = InvalidBucketError("Error message")
-    mock_get_parser.return_value = mock_parser
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 403
-    assert response.json() == {"detail": "Invalid bucket specified."}
-
-
-def test_save_recording_filetype_error(recording_settings, mock_get_parser):
-    """Test handling of unsupported file types in recording event data."""
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.side_effect = InvalidFileTypeError(
-        "unsupported '.json'"
-    )
-    mock_get_parser.return_value = mock_parser
-
-    client = APIClient()
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 200
-    assert response.json() == {"message": "Notification ignored."}
-
-
-def test_save_recording_filepath_error(recording_settings, mock_get_parser):
-    """Test handling of unsupported filepath in recording event data."""
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.side_effect = InvalidFilepathError(
-        "Invalid filepath structure: parent/folder/recording.jpeg"
-    )
-    mock_get_parser.return_value = mock_parser
-
-    client = APIClient()
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 200
-    assert response.json() == {"message": "Notification ignored."}
-
-
-def test_save_recording_unknown_recording(recording_settings, mock_get_parser, client):
-    """Test handling of events for non-existent recordings."""
-
-    RecordingFactory(status="active")
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.return_value = uuid.uuid4()
-    mock_get_parser.return_value = mock_parser
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 404
-    assert response.json() == {"detail": "No recording found for this event."}
-
-
-@pytest.mark.parametrize(
-    "status", ["failed_to_start", "aborted", "failed_to_stop", "saved", "initiated"]
-)
-def test_save_recording_non_savable_recording(
-    recording_settings, mock_get_parser, client, status
-):
-    """Test that recordings in non-savable states cannot be saved."""
-
-    recording = RecordingFactory(status=status)
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.return_value = recording.id
-    mock_get_parser.return_value = mock_parser
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 403
-    assert response.json() == {
-        "detail": f"Recording with ID {recording.id} cannot be saved because it is either,"
-        " in an error state or has already been saved."
-    }
-
-
-@pytest.mark.parametrize("status", ["active", "stopped"])
-def test_save_recording_success(recording_settings, mock_get_parser, client, status):
-    """Test successful saving of recordings in valid states."""
-
-    recording = RecordingFactory(status=status)
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.return_value = recording.id
-    mock_get_parser.return_value = mock_parser
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 200
-    assert response.json() == {"message": "Event processed."}
-
-    recording.refresh_from_db()
-    assert recording.status == RecordingStatusChoices.SAVED
-
-
-@mock.patch(
-    "core.recording.services.recording_events.notification_service."
-    "notify_external_services"
-)
-@pytest.mark.parametrize("notification_succeeded", [True, False])
-def test_save_recording_notifies_external_services(
-    mock_notify_external_services,
-    recording_settings,
-    mock_get_parser,
-    client,
-    notification_succeeded,
-):
-    """External services should be notified when a recording is saved."""
-
-    recording = RecordingFactory(status="active")
-
-    mock_parser = mock.Mock()
-    mock_parser.get_recording_id.return_value = recording.id
-    mock_get_parser.return_value = mock_parser
-
-    mock_notify_external_services.return_value = notification_succeeded
-
-    response = client.post(
-        "/api/v1.0/recordings/storage-hook/",
-        {"recording_data": "valid-data"},
-        HTTP_AUTHORIZATION="Bearer testAuthToken",
-    )
-
-    assert response.status_code == 200
-    assert response.json() == {"message": "Event processed."}
-
-    mock_notify_external_services.assert_called_once_with(recording)
-
-    recording.refresh_from_db()
-    assert recording.status == (
-        RecordingStatusChoices.NOTIFICATION_SUCCEEDED
-        if notification_succeeded
-        else RecordingStatusChoices.SAVED
-    )

src/backend/core/tests/services/test_livekit_events.py

@@ -75,12 +75,11 @@ def test_initialization(
 )
 @mock.patch("core.utils.notify_participants")
 @mock.patch("core.services.room_management.RoomManagement.update_metadata")
-def test_handle_egress_ended_success(  # noqa: PLR0913, PLR0917 # pylint: disable=too-many-arguments, too-many-positional-arguments
-    mock_update_metadata, mock_notify, mode, notification_type, service, settings
+def test_handle_egress_ended_success(  # pylint: disable=too-many-arguments, too-many-positional-arguments
+    mock_update_metadata, mock_notify, mode, notification_type, service
 ):
     """Should successfully stop recording and notifies all participant."""
 
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
     recording = RecordingFactory(worker_id="worker-1", mode=mode, status="active")
     mock_data = mock.MagicMock()
     mock_data.egress_info.egress_id = recording.worker_id
@@ -159,12 +158,11 @@ def test_handle_egress_updated_non_handled(
 )
 @mock.patch("core.utils.notify_participants")
 @mock.patch("core.services.room_management.RoomManagement.update_metadata")
-def test_handle_egress_ended_metadata_update_fails(  # noqa: PLR0913, PLR0917 # pylint: disable=too-many-arguments, too-many-positional-arguments
-    mock_update_metadata, mock_notify, mode, notification_type, service, settings
+def test_handle_egress_ended_metadata_update_fails(  # pylint: disable=too-many-arguments, too-many-positional-arguments
+    mock_update_metadata, mock_notify, mode, notification_type, service
 ):
     """Should successfully stop and save recording when metadata's update fails."""
 
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
     recording = RecordingFactory(worker_id="worker-1", mode=mode, status="active")
     mock_data = mock.MagicMock()
     mock_data.egress_info.egress_id = recording.worker_id
@@ -358,12 +356,10 @@ def test_handle_egress_ended_finalizes_recording(  # noqa: PLR0913, PLR0917
     recording_status,
     egress_status,
     service,
-    settings,
 ):  # pylint: disable=too-many-arguments,too-many-positional-arguments
     """Should notify external services and save the recording on egress completion
-    (EGRESS_COMPLETE or EGRESS_LIMIT_REACHED) when RECORDING_STORAGE_EVENT_ENABLE is False.
+    (EGRESS_COMPLETE or EGRESS_LIMIT_REACHED).
     """
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
     mock_notify_external_services.return_value = notify_return_value
 
     recording = RecordingFactory(worker_id="worker-1", status="active")
@@ -379,47 +375,6 @@ def test_handle_egress_ended_finalizes_recording(  # noqa: PLR0913, PLR0917
     assert recording.status == recording_status
 
 
-@mock.patch(
-    "core.recording.services.recording_events.notification_service."
-    "notify_external_services"
-)
-@mock.patch("core.utils.notify_participants")
-@mock.patch("core.services.room_management.RoomManagement.update_metadata")
-@pytest.mark.parametrize(
-    "egress_status, expected_status",
-    [
-        (EgressStatus.EGRESS_COMPLETE, "active"),
-        (EgressStatus.EGRESS_LIMIT_REACHED, "stopped"),
-    ],
-)
-def test_handle_egress_ended_does_not_finalize_when_webhooks_enabled(  # noqa: PLR0913, PLR0917
-    mock_update_metadata,
-    mock_notify,
-    mock_notify_external_services,
-    egress_status,
-    expected_status,
-    service,
-    settings,
-):  # pylint: disable=too-many-arguments,too-many-positional-arguments
-    """When storage event webhooks are enabled, egress_ended must not finalize the
-    recording: external services are never notified. EGRESS_LIMIT_REACHED still stops
-    the recording, EGRESS_COMPLETE leaves it active.
-    """
-    settings.RECORDING_STORAGE_EVENT_ENABLE = True
-
-    recording = RecordingFactory(worker_id="worker-1", status="active")
-    mock_data = mock.MagicMock()
-    mock_data.egress_info.egress_id = recording.worker_id
-    mock_data.egress_info.status = egress_status
-
-    service._handle_egress_ended(mock_data)
-
-    mock_notify_external_services.assert_not_called()
-
-    recording.refresh_from_db()
-    assert recording.status == expected_status
-
-
 @pytest.mark.parametrize(
     "egress_status",
     [
@@ -432,10 +387,9 @@ def test_handle_egress_ended_does_not_finalize_when_webhooks_enabled(  # noqa: P
 )
 @mock.patch("core.services.room_management.RoomManagement.update_metadata")
 def test_handle_egress_ended_does_not_save_on_wrong_status(
-    mock_update_metadata, egress_status, service, settings
+    mock_update_metadata, egress_status, service
 ):
     """Shouldn't save on invalid status."""
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
 
     recording = RecordingFactory(worker_id="worker-1", status="active")
     mock_data = mock.MagicMock()
@@ -453,14 +407,13 @@ def test_handle_egress_ended_does_not_save_on_wrong_status(
 )
 @mock.patch("core.services.room_management.RoomManagement.update_metadata")
 def test_handle_egress_ended_ignores_non_savable_recording(
-    mock_update_metadata, status, service, settings
+    mock_update_metadata, status, service
 ):
     """Should handle non-savable recordings idempotently without raising.
 
     'egress_ended' may be redelivered (e.g. for an already-saved recording);
     this must not raise, otherwise the webhook would 500 and LiveKit would retry.
     """
-    settings.RECORDING_STORAGE_EVENT_ENABLE = False
 
     recording = RecordingFactory(worker_id="worker-1", status=status)
     mock_data = mock.MagicMock()

src/backend/meet/settings.py

@@ -729,20 +729,6 @@ class Base(Configuration):
         environ_name="RECORDING_WORKER_CLASSES",
         environ_prefix=None,
     )
-    RECORDING_EVENT_PARSER_CLASS = values.Value(
-        "core.recording.event.parsers.MinioParser",
-        environ_name="RECORDING_EVENT_PARSER_CLASS",
-        environ_prefix=None,
-    )
-    RECORDING_ENABLE_STORAGE_EVENT_AUTH = values.BooleanValue(
-        True, environ_name="RECORDING_ENABLE_STORAGE_EVENT_AUTH", environ_prefix=None
-    )
-    RECORDING_STORAGE_EVENT_ENABLE = values.BooleanValue(
-        False, environ_name="RECORDING_STORAGE_EVENT_ENABLE", environ_prefix=None
-    )
-    RECORDING_STORAGE_EVENT_TOKEN = SecretFileValue(
-        None, environ_name="RECORDING_STORAGE_EVENT_TOKEN", environ_prefix=None
-    )
     # Number of days before recordings expire - must be synced with bucket lifecycle policy
     # Set to None for no expiration
     RECORDING_EXPIRATION_DAYS = values.IntegerValue(

src/frontend/src/features/devtools/MeetDevtools.tsx

@@ -0,0 +1,608 @@
+import { CSSProperties, useState } from 'react'
+import { useConnectionState, useRoomContext } from '@livekit/components-react'
+import { ConnectionState } from 'livekit-client'
+import { StatsSnapshot, TrackRow, useWebRTCStats } from './useWebRTCStats'
+import { readRoomConfig } from './roomConfig'
+import {
+  forceTransport,
+  releaseForcedTransport,
+  Scenario,
+  SCENARIOS,
+  setDownlinkCap,
+  setUplinkCap,
+  stepBitrate,
+  TransportMode,
+  transportModeFromRoute,
+} from './simulation'
+
+const SANS =
+  "-apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif"
+const MONO = 'ui-monospace, SFMono-Regular, Menlo, Consolas, monospace'
+
+const COLOR = {
+  text: '#e8e8e8',
+  muted: '#9a9a9a',
+  faint: '#6f6f6f',
+  hairline: '#2c2d31',
+  border: '#3d3e44',
+  surface: '#1b1c1e',
+  chartBg: '#141517',
+  down: '#f6821f',
+  up: '#5a9cf8',
+  ok: '#10b981',
+  bad: '#f05a4a',
+  busy: '#e5a13c',
+}
+
+const stateColor = (state: ConnectionState): string => {
+  switch (state) {
+    case ConnectionState.Connected:
+      return COLOR.ok
+    case ConnectionState.Reconnecting:
+    case ConnectionState.SignalReconnecting:
+      return COLOR.busy
+    case ConnectionState.Connecting:
+      return COLOR.up
+    case ConnectionState.Disconnected:
+      return COLOR.bad
+  }
+}
+
+const styles: Record<string, CSSProperties> = {
+  toggle: {
+    position: 'fixed',
+    bottom: 12,
+    right: 12,
+    zIndex: 9999,
+    fontFamily: SANS,
+    fontSize: 12,
+    lineHeight: 1,
+    padding: '8px 12px',
+    borderRadius: 999,
+    borderWidth: 1,
+    borderStyle: 'solid',
+    borderColor: COLOR.border,
+    background: COLOR.surface,
+    color: COLOR.text,
+    cursor: 'pointer',
+    boxShadow: '0 2px 10px #0006',
+  },
+  panel: {
+    position: 'fixed',
+    bottom: 12,
+    right: 12,
+    zIndex: 9999,
+    width: 460,
+    maxWidth: 'calc(100vw - 24px)',
+    maxHeight: 'calc(100vh - 24px)',
+    overflowY: 'auto',
+    fontFamily: SANS,
+    fontSize: 11,
+    lineHeight: 1.5,
+    color: COLOR.text,
+    background: COLOR.surface,
+    borderWidth: 1,
+    borderStyle: 'solid',
+    borderColor: COLOR.border,
+    borderRadius: 10,
+    padding: 14,
+    boxShadow: '0 10px 34px #00000080',
+  },
+  sectionTitle: {
+    margin: '14px 0 5px',
+    color: COLOR.faint,
+    textTransform: 'uppercase',
+    letterSpacing: 1.2,
+    fontSize: 9,
+    fontWeight: 600,
+    display: 'flex',
+    justifyContent: 'space-between',
+  },
+  metricsLine: {
+    display: 'flex',
+    alignItems: 'center',
+    gap: 16,
+    margin: '0 0 10px',
+    paddingBottom: 10,
+    borderBottom: `1px solid ${COLOR.hairline}`,
+    fontVariantNumeric: 'tabular-nums',
+  },
+  legend: {
+    display: 'flex',
+    gap: 12,
+    marginTop: 3,
+    color: COLOR.faint,
+    fontSize: 10,
+  },
+  list: { display: 'flex', flexDirection: 'column', maxHeight: '200px', overflowY: 'auto' },
+  listRow: { display: 'flex', gap: 8, padding: '1px 0' },
+  listLabel: { flex: '0 0 34%', color: COLOR.muted },
+  trackRow: {
+    display: 'flex',
+    gap: 8,
+    padding: '3px 0',
+    borderBottom: `1px solid ${COLOR.hairline}`,
+    alignItems: 'baseline',
+  },
+  trackLabel: {
+    flex: '0 0 34%',
+    whiteSpace: 'nowrap',
+    overflow: 'hidden',
+    textOverflow: 'ellipsis',
+  },
+  trackCodec: {
+    flex: '0 0 11%',
+    fontFamily: MONO,
+    fontSize: 10,
+    color: COLOR.muted,
+  },
+  trackKbps: {
+    flex: '0 0 9%',
+    textAlign: 'right',
+    fontVariantNumeric: 'tabular-nums',
+  },
+  trackDetail: { flex: 1, minWidth: 0 },
+  mono: { fontFamily: MONO, fontSize: 10, color: COLOR.muted },
+  button: {
+    fontFamily: SANS,
+    fontSize: 11,
+    padding: '3px 9px',
+    borderRadius: 5,
+    borderWidth: 1,
+    borderStyle: 'solid',
+    borderColor: COLOR.border,
+    background: '#232428',
+    color: COLOR.text,
+    cursor: 'pointer',
+  },
+  buttonActive: { borderColor: COLOR.down, color: COLOR.down },
+  buttonDisabled: { color: COLOR.faint, cursor: 'not-allowed' },
+  row: { display: 'flex', alignItems: 'center', gap: 6, flexWrap: 'wrap' },
+  stepperValue: {
+    minWidth: 92,
+    textAlign: 'center',
+    borderWidth: 1,
+    borderStyle: 'solid',
+    borderColor: COLOR.hairline,
+    borderRadius: 4,
+    padding: '2px 6px',
+    background: COLOR.chartBg,
+    fontVariantNumeric: 'tabular-nums',
+  },
+  close: {
+    background: 'none',
+    border: 'none',
+    color: COLOR.faint,
+    cursor: 'pointer',
+    fontSize: 14,
+  },
+}
+
+const Sparkline = ({ history }: { history: StatsSnapshot[] }) => {
+  const width = 430
+  const height = 54
+  const max = Math.max(...history.map((s) => Math.max(s.upKbps, s.downKbps)), 1)
+  const points = (pick: (s: StatsSnapshot) => number) =>
+    history
+      .map(
+        (s, i) =>
+          `${((i / Math.max(history.length - 1, 1)) * width).toFixed(1)},${(
+            height -
+            (pick(s) / max) * (height - 6) -
+            3
+          ).toFixed(1)}`
+      )
+      .join(' ')
+  return (
+    <>
+      <svg
+        width={width}
+        height={height}
+        role="img"
+        aria-label="Bandwidth over the last minute"
+        style={{
+          display: 'block',
+          background: COLOR.chartBg,
+          borderRadius: 6,
+          border: `1px solid ${COLOR.hairline}`,
+        }}
+      >
+        {history.length >= 2 && (
+          <>
+            <polyline
+              points={points((s) => s.downKbps)}
+              fill="none"
+              stroke={COLOR.down}
+              strokeWidth="1.5"
+            />
+            <polyline
+              points={points((s) => s.upKbps)}
+              fill="none"
+              stroke={COLOR.up}
+              strokeWidth="1.5"
+            />
+          </>
+        )}
+      </svg>
+      {/* Legend outside the plot: text over the lines was unreadable. */}
+      <div style={styles.legend}>
+        <span>
+          <span style={{ color: COLOR.down }}>●</span> down
+        </span>
+        <span>
+          <span style={{ color: COLOR.up }}>●</span> up
+        </span>
+        <span>last 60s · max {Math.round(max)} kbps</span>
+      </div>
+    </>
+  )
+}
+
+const TrackDetail = ({ track }: { track: TrackRow }) => {
+  const parts: Array<{ text: string; tone?: 'warn' | 'muted' }> = []
+  if (track.res) parts.push({ text: track.res })
+  if (track.fps !== undefined) {
+    parts.push({ text: `${Math.round(track.fps)}fps` })
+  }
+  if (track.lossPct !== undefined && track.lossPct > 0) {
+    parts.push({ text: `loss ${track.lossPct.toFixed(1)}%`, tone: 'warn' })
+  }
+  if (track.froze) parts.push({ text: 'freeze', tone: 'warn' })
+  if (track.limitation) {
+    parts.push({ text: `lim:${track.limitation}`, tone: 'warn' })
+  }
+  const layer = track.layer
+  if (layer) {
+    if (layer.availRes && layer.availRes !== track.res) {
+      parts.push({
+        text: `≤${layer.availRes}${layer.layerCount ? `(${layer.layerCount}L)` : ''}`,
+        tone: 'muted',
+      })
+    }
+    if (layer.reason === 'adaptive' && layer.elementRes) {
+      parts.push({ text: `fit:${layer.elementRes}`, tone: 'muted' })
+    } else if (layer.reason === 'bandwidth') {
+      parts.push({ text: 'bw-limited', tone: 'warn' })
+    } else if (layer.reason === 'paused') {
+      parts.push({ text: 'paused:SFU', tone: 'warn' })
+    } else if (layer.reason === 'off-screen') {
+      parts.push({ text: 'off-screen', tone: 'muted' })
+    }
+  }
+  return (
+    <>
+      {parts.map((part, i) => (
+        <span
+          key={part.text}
+          style={
+            part.tone === 'warn'
+              ? { color: COLOR.bad }
+              : part.tone === 'muted'
+                ? { color: COLOR.faint }
+                : {}
+          }
+        >
+          {i > 0 && ' · '}
+          {part.text}
+        </span>
+      ))}
+    </>
+  )
+}
+
+const CapStepper = ({
+  label,
+  hint,
+  valueKbps,
+  onChange,
+}: {
+  label: string
+  hint: string
+  valueKbps: number | null
+  onChange: (kbps: number | null) => void
+}) => (
+  <div style={styles.row}>
+    <span style={{ flex: '0 0 70px' }}>{label}</span>
+    <button
+      type="button"
+      style={styles.button}
+      aria-label={`Decrease ${label} bandwidth`}
+      onClick={() => onChange(stepBitrate(valueKbps, 'decrease'))}
+    >
+      −
+    </button>
+    <span
+      style={{
+        ...styles.stepperValue,
+        ...(valueKbps !== null ? { color: COLOR.down } : {}),
+      }}
+    >
+      {valueKbps === null ? 'unlimited' : `≤ ${valueKbps} kbps`}
+    </span>
+    <button
+      type="button"
+      style={styles.button}
+      aria-label={`Increase ${label} bandwidth`}
+      onClick={() => onChange(stepBitrate(valueKbps, 'increase'))}
+    >
+      +
+    </button>
+    <span style={styles.mono}>{hint}</span>
+  </div>
+)
+
+const TrackList = ({ title, rows }: { title: string; rows: TrackRow[] }) => (
+  <>
+    <div style={styles.sectionTitle}>
+      <span>
+        {title} ({rows.length})
+      </span>
+      <span>
+        {Math.round(rows.reduce((sum, t) => sum + t.kbps, 0))} kbps media
+      </span>
+    </div>
+    <div style={styles.list}>
+      {rows.map((track) => (
+        <div key={track.key} style={styles.trackRow}>
+          <span style={styles.trackLabel}>{track.label}</span>
+          <span style={styles.trackCodec}>{track.codec ?? '–'}</span>
+          <span style={styles.trackKbps}>{Math.round(track.kbps)}</span>
+          <span style={styles.trackDetail}>
+            <TrackDetail track={track} />
+          </span>
+        </div>
+      ))}
+      {rows.length === 0 && <span style={{ color: COLOR.faint }}>none</span>}
+    </div>
+  </>
+)
+
+const MeetDevtools = () => {
+  const room = useRoomContext()
+  const connState = useConnectionState(room)
+  const [open, setOpen] = useState(false)
+  const [firedScenario, setFiredScenario] = useState<string>()
+  const [uplinkCap, setUplinkCapState] = useState<number | null>(null)
+  const [downlinkCap, setDownlinkCapState] = useState<number | null>(null)
+  const { snapshot, history } = useWebRTCStats(room, open)
+
+  const transportMode: TransportMode = transportModeFromRoute(snapshot?.route)
+
+  const fireScenario = (scenario: Scenario) => {
+    setFiredScenario(scenario.id)
+    scenario.run(room).catch((e) => {
+      console.warn('[MeetDevtools] simulateScenario failed', e)
+    })
+    window.setTimeout(
+      () =>
+        setFiredScenario((current) =>
+          current === scenario.id ? undefined : current
+        ),
+      1200
+    )
+  }
+
+  if (!open) {
+    return (
+      <button
+        type="button"
+        style={styles.toggle}
+        onClick={() => setOpen(true)}
+        aria-label="Open WebRTC devtools"
+      >
+        <span style={{ color: stateColor(connState) }}>●</span> rtc
+      </button>
+    )
+  }
+
+  const config = readRoomConfig(room)
+  const published = snapshot?.tracks.filter((t) => t.dir === 'up') ?? []
+  const subscribed = snapshot?.tracks.filter((t) => t.dir === 'down') ?? []
+  const turnProtocols = snapshot?.turnProtocols ?? []
+  const transports: Array<{
+    mode: TransportMode
+    label: string
+    requires?: string
+    title: string
+  }> = [
+    {
+      mode: 'auto',
+      label: 'auto (udp)',
+      title:
+        'clears the server-cached transport preference and relay-only policy, then full reconnect',
+    },
+    {
+      mode: 'tcp',
+      label: 'tcp',
+      title: 'force-tcp — server prefers TCP candidates (ICE/TCP)',
+    },
+    {
+      mode: 'turn-udp',
+      label: 'turn:udp',
+      requires: 'udp',
+      title:
+        'client-side: iceTransportPolicy relay + full reconnect (no server hook exists)',
+    },
+    {
+      mode: 'turn-tcp',
+      label: 'turn:tcp',
+      requires: 'tcp',
+      title: 'force-tcp — lands on TURN/TCP when it is the TCP path',
+    },
+    {
+      mode: 'turn-tls',
+      label: 'turn:tls',
+      requires: 'tls',
+      title: 'force-tls — server switches you to TURN over TLS',
+    },
+  ]
+
+  return (
+    <section style={styles.panel} aria-label="WebRTC devtools">
+      <div style={styles.metricsLine}>
+        <span>
+          <span style={{ color: COLOR.down }}>↓</span> {snapshot?.downKbps ?? 0}{' '}
+          kbps
+        </span>
+        <span>
+          <span style={{ color: COLOR.up }}>↑</span> {snapshot?.upKbps ?? 0}{' '}
+          kbps
+        </span>
+        <span style={{ color: COLOR.muted }}>
+          rtt{' '}
+          <span style={{ color: COLOR.text }}>{snapshot?.rttMs ?? '–'}</span> ms
+        </span>
+        <span style={{ color: COLOR.muted }}>
+          jitter{' '}
+          <span style={{ color: COLOR.text }}>{snapshot?.jitterMs ?? '–'}</span>{' '}
+          ms
+        </span>
+        <span
+          style={{
+            color: stateColor(connState),
+            fontSize: 10,
+            marginLeft: 'auto',
+          }}
+        >
+          ● {connState}
+        </span>
+        <button
+          type="button"
+          style={styles.close}
+          onClick={() => setOpen(false)}
+          aria-label="Close WebRTC devtools"
+        >
+          ✕
+        </button>
+      </div>
+
+      <Sparkline history={history} />
+
+      <div style={styles.sectionTitle}>room configuration (live)</div>
+      <div style={styles.list}>
+        {config.map((entry) => (
+          <div key={entry.label} style={styles.listRow}>
+            <span
+              style={{
+                color:
+                  entry.on === undefined
+                    ? COLOR.faint
+                    : entry.on
+                      ? COLOR.ok
+                      : COLOR.bad,
+              }}
+            >
+              ●
+            </span>
+            <span style={styles.listLabel}>{entry.label}</span>
+            <span>{entry.value}</span>
+          </div>
+        ))}
+      </div>
+
+      <TrackList title="published ↑" rows={published} />
+      <TrackList title="subscribed ↓" rows={subscribed} />
+
+      <div style={styles.sectionTitle}>bandwidth</div>
+      <div style={{ ...styles.list, gap: 4 }}>
+        <CapStepper
+          label="uplink"
+          hint="encoder cap (setParameters)"
+          valueKbps={uplinkCap}
+          onChange={(kbps) => {
+            setUplinkCapState(kbps)
+            void setUplinkCap(room, kbps)
+          }}
+        />
+        <CapStepper
+          label="downlink"
+          hint="SFU limit (subscriber-bandwidth)"
+          valueKbps={downlinkCap}
+          onChange={(kbps) => {
+            setDownlinkCapState(kbps)
+            void setDownlinkCap(room, kbps).catch((e) =>
+              console.warn('[MeetDevtools] downlink cap failed', e)
+            )
+          }}
+        />
+      </div>
+
+      <div style={styles.sectionTitle}>transport</div>
+      <div style={styles.row}>
+        {transports.map(({ mode, label, requires, title }) => {
+          const available = !requires || turnProtocols.includes(requires)
+          const active = transportMode === mode
+          const clickable = available && !active
+          return (
+            <button
+              key={mode}
+              type="button"
+              aria-pressed={active}
+              disabled={!clickable}
+              title={available ? title : `${title} — not configured`}
+              style={{
+                ...styles.button,
+                ...(clickable || active ? {} : styles.buttonDisabled),
+                ...(active ? styles.buttonActive : {}),
+              }}
+              onClick={() => {
+                if (!clickable) return
+                void (
+                  mode === 'auto'
+                    ? releaseForcedTransport(room)
+                    : forceTransport(
+                        room,
+                        mode as 'tcp' | 'turn-udp' | 'turn-tcp' | 'turn-tls'
+                      )
+                ).catch((e) =>
+                  console.warn('[MeetDevtools] transport change failed', e)
+                )
+              }}
+            >
+              {requires && (
+                <span style={{ color: available ? COLOR.ok : COLOR.faint }}>
+                  ●{' '}
+                </span>
+              )}
+              {label}
+            </button>
+          )
+        })}
+        <span style={styles.mono}>
+          route:{' '}
+          {snapshot?.route
+            ? [
+                snapshot.route.protocol,
+                snapshot.route.type,
+                snapshot.route.relayProtocol &&
+                  `relay:${snapshot.route.relayProtocol}`,
+              ]
+                .filter(Boolean)
+                .join('·')
+            : '–'}
+        </span>
+      </div>
+
+      <div style={styles.sectionTitle}>connection scenarios</div>
+      <div style={styles.row}>
+        {SCENARIOS.map((scenario) => (
+          <button
+            key={scenario.id}
+            type="button"
+            title={`${scenario.side}-side simulation — momentary, watch the state dot`}
+            style={{
+              ...styles.button,
+              ...(firedScenario === scenario.id ? styles.buttonActive : {}),
+            }}
+            onClick={() => fireScenario(scenario)}
+          >
+            {scenario.side === 'server' ? '☁ ' : ''}
+            {scenario.label}
+          </button>
+        ))}
+      </div>
+    </section>
+  )
+}
+
+export default MeetDevtools

src/frontend/src/features/devtools/index.tsx

@@ -0,0 +1,15 @@
+import { lazy, Suspense } from 'react'
+
+
+const LazyPanel = import.meta.env.DEV
+  ? lazy(() => import('./MeetDevtools'))
+  : null
+
+export const MeetDevtools = () => {
+  if (!LazyPanel) return null
+  return (
+    <Suspense fallback={null}>
+      <LazyPanel />
+    </Suspense>
+  )
+}

src/frontend/src/features/devtools/roomConfig.ts

@@ -0,0 +1,89 @@
+import { Room } from 'livekit-client'
+
+
+export type ConfigEntry = {
+  label: string
+  value: string
+  /** true = feature actively on, false = off, undefined = informational */
+  on?: boolean
+}
+
+const formatBackupCodec = (backup: unknown): string => {
+  if (backup === undefined || backup === true) return 'auto'
+  if (backup === false) return 'off'
+  if (typeof backup === 'object' && backup !== null && 'codec' in backup) {
+    return String((backup as { codec: unknown }).codec)
+  }
+  return String(backup)
+}
+
+export const readRoomConfig = (room: Room): ConfigEntry[] => {
+  const options = room.options
+  const publish = options.publishDefaults
+  const adaptive = options.adaptiveStream
+
+  const entries: ConfigEntry[] = [
+    {
+      label: 'adaptiveStream',
+      value:
+        typeof adaptive === 'object'
+          ? JSON.stringify(adaptive)
+          : String(!!adaptive),
+      on: !!adaptive,
+    },
+    {
+      label: 'dynacast',
+      value: String(!!options.dynacast),
+      on: !!options.dynacast,
+    },
+    {
+      label: 'e2ee',
+      value: String(room.isE2EEEnabled),
+      on: room.isE2EEEnabled,
+    },
+    {
+      label: 'videoCodec',
+      value: publish?.videoCodec ?? 'default',
+    },
+    {
+      label: 'backupCodec',
+      value: formatBackupCodec(publish?.backupCodec),
+    },
+    {
+      label: 'simulcast',
+      value: String(publish?.simulcast ?? true),
+      on: publish?.simulcast ?? true,
+    },
+    {
+      label: 'audio dtx',
+      value: String(publish?.dtx ?? true),
+      on: publish?.dtx ?? true,
+    },
+    {
+      label: 'audio red',
+      value: String(publish?.red ?? true),
+      on: publish?.red ?? true,
+    },
+    {
+      label: 'quality (local)',
+      value: room.localParticipant.connectionQuality,
+    },
+  ]
+
+  const server = room.serverInfo
+  if (server) {
+    entries.push({
+      label: 'server',
+      value: [
+        server.version && `v${server.version}`,
+        server.region,
+        server.protocol !== undefined && `proto ${server.protocol}`,
+        server.edition !== undefined && `edition ${server.edition}`,
+      ]
+        .filter(Boolean)
+        .join(' · '),
+    })
+  }
+
+  return entries
+}

src/frontend/src/features/devtools/simulation.ts

@@ -0,0 +1,193 @@
+import { Room, Track } from 'livekit-client'
+// Transitive dependency of livekit-client (pinned by it); used only to
+// build the one signal request room.simulateScenario cannot express.
+import { SimulateScenario } from '@livekit/protocol'
+
+/** Shared bandwidth ladder for the −/+ steppers; null = unlimited. */
+export const BITRATE_LADDER_KBPS: Array<number | null> = [
+  null,
+  2000,
+  1000,
+  600,
+  300,
+  150,
+]
+
+export const stepBitrate = (
+  current: number | null,
+  direction: 'decrease' | 'increase'
+): number | null => {
+  const index = BITRATE_LADDER_KBPS.indexOf(current)
+  const safeIndex = index === -1 ? 0 : index
+  const next =
+    direction === 'decrease'
+      ? Math.min(safeIndex + 1, BITRATE_LADDER_KBPS.length - 1)
+      : Math.max(safeIndex - 1, 0)
+  return BITRATE_LADDER_KBPS[next]
+}
+
+const savedEncodings = new WeakMap<RTCRtpSender, Array<number | undefined>>()
+
+export const setUplinkCap = async (
+  room: Room,
+  kbps: number | null
+): Promise<void> => {
+  const senders: RTCRtpSender[] = []
+  room.localParticipant.trackPublications.forEach((pub) => {
+    const track = pub.track
+    if (track?.kind === Track.Kind.Video && track.sender) {
+      senders.push(track.sender)
+    }
+  })
+
+  for (const sender of senders) {
+    const params = sender.getParameters()
+    if (!params.encodings || params.encodings.length === 0) continue
+
+    if (kbps === null) {
+      const original = savedEncodings.get(sender)
+      params.encodings.forEach((encoding, i) => {
+        encoding.maxBitrate = original?.[i]
+      })
+      savedEncodings.delete(sender)
+    } else {
+      if (!savedEncodings.has(sender)) {
+        savedEncodings.set(
+          sender,
+          params.encodings.map((encoding) => encoding.maxBitrate)
+        )
+      }
+      const activeCount =
+        params.encodings.filter((encoding) => encoding.active !== false)
+          .length || 1
+      // Split the budget across active simulcast layers / SVC encoding.
+      const perEncoding = Math.max(
+        30_000,
+        Math.floor((kbps * 1000) / activeCount)
+      )
+      params.encodings.forEach((encoding) => {
+        encoding.maxBitrate = perEncoding
+      })
+    }
+
+    try {
+      await sender.setParameters(params)
+    } catch (e) {
+      console.warn('[MeetDevtools] setParameters failed', e)
+    }
+  }
+}
+
+
+export const setDownlinkCap = (
+  room: Room,
+  kbps: number | null
+): Promise<void> =>
+  room.simulateScenario('subscriber-bandwidth', kbps === null ? 0 : kbps * 1000)
+
+export type Scenario = {
+  id: string
+  label: string
+  /** where the simulation happens */
+  side: 'client' | 'server'
+  run: (room: Room) => Promise<void>
+}
+
+export const SCENARIOS: Scenario[] = [
+  {
+    id: 'resume',
+    label: 'reconnect (resume)',
+    side: 'client',
+    // Replays a signaling WebSocket loss; media keeps flowing, client resumes.
+    run: (room) => room.simulateScenario('signal-reconnect'),
+  },
+  {
+    id: 'resume-fail',
+    label: 'reconnect (resume fails)',
+    side: 'client',
+    // Same, but the next resume attempt fails → exercises the retry ladder.
+    run: (room) => room.simulateScenario('resume-reconnect'),
+  },
+  {
+    id: 'full-reconnect',
+    label: 'full reconnect',
+    side: 'client',
+    // Complete rejoin with brand-new peer connections.
+    run: (room) => room.simulateScenario('full-reconnect'),
+  },
+  {
+    id: 'migration',
+    label: 'server migration',
+    side: 'server',
+    run: (room) => room.simulateScenario('migration'),
+  },
+  {
+    id: 'node-failure',
+    label: 'SFU node failure',
+    side: 'server',
+    run: (room) => room.simulateScenario('node-failure'),
+  },
+  {
+    id: 'server-leave',
+    label: 'server disconnect',
+    side: 'server',
+    // Server-initiated leave: the closest thing to "you got kicked".
+    run: (room) => room.simulateScenario('server-leave'),
+  },
+]
+
+export type TransportMode =
+  | 'auto'
+  | 'tcp'
+  | 'turn-udp'
+  | 'turn-tcp'
+  | 'turn-tls'
+
+const clearServerTransportPreference = (room: Room): Promise<void> =>
+  room.engine.client.sendSimulateScenario(
+    new SimulateScenario({
+      scenario: { case: 'switchCandidateProtocol', value: 0 },
+    })
+  )
+
+const setRelayOnly = (room: Room, relay: boolean) => {
+  room.engine.rtcConfig = {
+    ...room.engine.rtcConfig,
+    iceTransportPolicy: relay ? 'relay' : 'all',
+  }
+}
+
+const settle = () => new Promise((resolve) => setTimeout(resolve, 300))
+
+export const forceTransport = async (
+  room: Room,
+  mode: 'tcp' | 'turn-udp' | 'turn-tcp' | 'turn-tls'
+): Promise<void> => {
+  if (mode === 'turn-udp') {
+    await clearServerTransportPreference(room)
+    setRelayOnly(room, true)
+    await settle()
+    return room.simulateScenario('full-reconnect')
+  }
+  setRelayOnly(room, false)
+  return room.simulateScenario(mode === 'turn-tls' ? 'force-tls' : 'force-tcp')
+}
+
+export const releaseForcedTransport = async (room: Room): Promise<void> => {
+  setRelayOnly(room, false)
+  await clearServerTransportPreference(room)
+  await settle()
+  await room.simulateScenario('full-reconnect')
+}
+
+export const transportModeFromRoute = (route?: {
+  protocol?: string
+  relayProtocol?: string
+}): TransportMode => {
+  // relayProtocol is the client→TURN leg; protocol alone means no relay.
+  if (route?.relayProtocol === 'tls') return 'turn-tls'
+  if (route?.relayProtocol === 'tcp') return 'turn-tcp'
+  if (route?.relayProtocol === 'udp') return 'turn-udp'
+  if (route?.protocol === 'tcp') return 'tcp'
+  return 'auto'
+}

src/frontend/src/features/devtools/useWebRTCStats.ts

@@ -0,0 +1,357 @@
+import { useEffect, useRef, useState } from 'react'
+import { Room, Track } from 'livekit-client'
+
+
+export type LayerInfo = {
+  /** top published layer, e.g. '1280x720' */
+  availRes?: string
+  /** number of published spatial layers (simulcast/SVC), when known */
+  layerCount?: number
+  /** largest attached element size in device px, e.g. '480x270' */
+  elementRes?: string
+  /** why the forwarded layer is below the top one */
+  reason?: 'adaptive' | 'bandwidth' | 'paused' | 'off-screen'
+}
+
+export type TrackRow = {
+  key: string
+  dir: 'up' | 'down'
+  kind: string
+  label: string
+  codec?: string
+  kbps: number
+  fps?: number
+  res?: string
+  lossPct?: number
+  limitation?: string
+  /** subscribed video only: SFU layer forwarding context */
+  layer?: LayerInfo
+  /** this track froze during the last tick */
+  froze?: boolean
+}
+
+export type StatsSnapshot = {
+  ts: number
+  /** wire totals from transport stats (includes headers, RTCP, FEC) */
+  upKbps: number
+  downKbps: number
+  rttMs?: number
+  /** worst inbound RTP jitter across subscribed tracks */
+  jitterMs?: number
+  availableOutKbps?: number
+  /** selected ICE route of the publisher transport (measured, not assumed) */
+  route?: { protocol?: string; type?: string; relayProtocol?: string }
+  /**
+   * relayProtocol values of gathered relay local candidates — i.e. which
+   * client→TURN transports are actually configured (udp/tcp/tls). Empty
+   * when no TURN server is configured.
+   */
+  turnProtocols: string[]
+  tracks: TrackRow[]
+}
+
+type StatDict = Record<string, unknown>
+type Counters = Record<string, number>
+
+const asNumber = (v: unknown): number | undefined =>
+  typeof v === 'number' && Number.isFinite(v) ? v : undefined
+
+const asString = (v: unknown): string | undefined =>
+  typeof v === 'string' ? v : undefined
+
+const shortCodec = (mimeType?: string) =>
+  mimeType ? mimeType.replace(/^(audio|video)\//, '') : undefined
+
+type TrackContext = { label: string; layer?: LayerInfo; publishedH?: number }
+
+/**
+ * msTrackId -> label + subscribed-layer context, rebuilt on every tick.
+ * The "why is this tile blurry" answer needs SDK state (publication +
+ * element size) joined with getStats (forwarded resolution); this is the
+ * SDK-state half.
+ */
+const buildTrackContext = (room: Room) => {
+  const contexts = new Map<string, TrackContext>()
+  room.localParticipant.trackPublications.forEach((pub) => {
+    const id = pub.track?.mediaStreamTrack?.id
+    if (id) contexts.set(id, { label: `local ${pub.source}` })
+  })
+  const dpr = window.devicePixelRatio || 1
+  room.remoteParticipants.forEach((participant) => {
+    // Keep rows scannable: participant names capped at 10 chars.
+    const rawName = participant.name || participant.identity
+    const name = rawName.length > 15 ? `${rawName.slice(0, 15)}.` : rawName
+    participant.trackPublications.forEach((pub) => {
+      const track = pub.track
+      const id = track?.mediaStreamTrack?.id
+      if (!track || !id) return
+      const label = `${name} ${pub.source}`
+      if (track.kind !== Track.Kind.Video) {
+        contexts.set(id, { label })
+        return
+      }
+      let elementRes: string | undefined
+      let elementH: number | undefined
+      for (const element of track.attachedElements) {
+        const w = Math.round(element.clientWidth * dpr)
+        const h = Math.round(element.clientHeight * dpr)
+        if (elementH === undefined || h > elementH) {
+          elementH = h
+          elementRes = `${w}x${h}`
+        }
+      }
+      const dims = pub.dimensions
+      const layers = pub.trackInfo?.layers?.length
+      contexts.set(id, {
+        label,
+        publishedH: dims?.height,
+        layer: {
+          availRes: dims ? `${dims.width}x${dims.height}` : undefined,
+          layerCount: layers && layers > 1 ? layers : undefined,
+          elementRes,
+          reason: !pub.isEnabled
+            ? 'off-screen'
+            : track.streamState === Track.StreamState.Paused
+              ? 'paused'
+              : undefined,
+        },
+      })
+    })
+  })
+  return contexts
+}
+
+/** Decide why a forwarded layer is below the published top layer. */
+const resolveLayerReason = (
+  context: TrackContext,
+  forwardedH: number | undefined
+): LayerInfo | undefined => {
+  const layer = context.layer
+  if (!layer) return undefined
+  if (layer.reason) return layer // off-screen / paused already decided
+  const publishedH = context.publishedH
+  if (!publishedH || !forwardedH || forwardedH >= publishedH * 0.9) {
+    return { ...layer, reason: undefined } // full quality, nothing to explain
+  }
+  // Below top layer: if the element only needs about what we get, it's
+  // adaptiveStream fitting the element; otherwise the SFU is holding back
+  // a layer the element could use — congestion.
+  const elementH = layer.elementRes
+    ? Number(layer.elementRes.split('x')[1])
+    : undefined
+  const adaptive = elementH !== undefined && forwardedH >= elementH * 0.7
+  return { ...layer, reason: adaptive ? 'adaptive' : 'bandwidth' }
+}
+
+export const useWebRTCStats = (
+  room: Room,
+  enabled: boolean,
+  intervalMs = 1000
+) => {
+  // Single source of truth: the snapshot is just the last history entry.
+  const [history, setHistory] = useState<StatsSnapshot[]>([])
+  const prevRef = useRef(new Map<string, Counters>())
+
+  useEffect(() => {
+    if (!enabled) return
+
+    let cancelled = false
+    const prev = prevRef.current
+
+    /** Per-stat counter deltas; returns 0 on the first sighting. */
+    const deltas = (key: string, now: Counters): Counters => {
+      const before = prev.get(key)
+      prev.set(key, now)
+      const out: Counters = {}
+      for (const [name, value] of Object.entries(now)) {
+        out[name] = before?.[name] !== undefined ? value - before[name] : 0
+      }
+      return out
+    }
+
+    const collect = async () => {
+      const reports: Array<{ pc: 'pub' | 'sub'; report: RTCStatsReport }> = []
+      try {
+        // Not public API — see file header.
+        const manager = room.engine?.pcManager
+        const pub = await manager?.publisher?.getStats()
+        if (pub) reports.push({ pc: 'pub', report: pub })
+        const sub = await manager?.subscriber?.getStats()
+        if (sub) reports.push({ pc: 'sub', report: sub })
+      } catch {
+        // Engine not ready or SDK internals changed; panel shows nothing.
+      }
+      if (cancelled || reports.length === 0) return
+
+      const contexts = buildTrackContext(room)
+      const tracks: TrackRow[] = []
+      let upKbps = 0
+      let downKbps = 0
+      let rttMs: number | undefined
+      let jitterMs: number | undefined
+      let availableOutKbps: number | undefined
+      let route: StatsSnapshot['route']
+      const turnProtocols = new Set<string>()
+
+      for (const { pc, report } of reports) {
+        const byId = new Map<string, StatDict>()
+        report.forEach((stat) => byId.set(stat.id as string, stat as StatDict))
+
+        report.forEach((raw) => {
+          const stat = raw as StatDict
+          const type = asString(stat.type)
+          const ts = asNumber(stat.timestamp) ?? Date.now()
+          const key = `${pc}:${asString(stat.id) ?? ''}`
+
+          if (
+            type === 'local-candidate' &&
+            asString(stat.candidateType) === 'relay'
+          ) {
+            // Relay candidates are only gathered when a TURN server is
+            // configured and reachable; relayProtocol says how the client
+            // reaches it (udp/tcp/tls) — a turn:…?transport=udp server
+            // must NOT be presented as a TURN/TLS capability.
+            turnProtocols.add(
+              asString(stat.relayProtocol) ?? asString(stat.protocol) ?? 'udp'
+            )
+          }
+
+          if (type === 'transport') {
+            const d = deltas(key, {
+              sent: asNumber(stat.bytesSent) ?? 0,
+              received: asNumber(stat.bytesReceived) ?? 0,
+              ts,
+            })
+            if (d.ts > 0) {
+              upKbps += Math.max(0, (d.sent * 8) / d.ts)
+              downKbps += Math.max(0, (d.received * 8) / d.ts)
+            }
+          }
+
+          if (type === 'candidate-pair' && stat.nominated === true) {
+            const rtt = asNumber(stat.currentRoundTripTime)
+            if (rtt !== undefined) rttMs = Math.round(rtt * 1000)
+            const available = asNumber(stat.availableOutgoingBitrate)
+            if (available !== undefined && pc === 'pub') {
+              availableOutKbps = Math.round(available / 1000)
+            }
+            if (pc === 'pub') {
+              const local = byId.get(asString(stat.localCandidateId) ?? '')
+              route = {
+                protocol: asString(local?.protocol),
+                type: asString(local?.candidateType),
+                relayProtocol: asString(local?.relayProtocol),
+              }
+            }
+          }
+
+          if (type === 'outbound-rtp' || type === 'inbound-rtp') {
+            const isUp = type === 'outbound-rtp'
+            const kind = asString(stat.kind)
+
+            if (!isUp) {
+              const jitter = asNumber(stat.jitter)
+              if (jitter !== undefined) {
+                const ms = Math.round(jitter * 1000)
+                if (jitterMs === undefined || ms > jitterMs) jitterMs = ms
+              }
+            }
+
+            // Packet loss: reported directly on inbound-rtp; for outbound it
+            // lives on the matching remote-inbound-rtp (what the SFU got).
+            let packetsLost = asNumber(stat.packetsLost) ?? 0
+            if (isUp) {
+              const remote = byId.get(asString(stat.remoteId) ?? '')
+              packetsLost = asNumber(remote?.packetsLost) ?? 0
+            }
+
+            const d = deltas(key, {
+              bytes: asNumber(isUp ? stat.bytesSent : stat.bytesReceived) ?? 0,
+              packets:
+                asNumber(isUp ? stat.packetsSent : stat.packetsReceived) ?? 0,
+              packetsLost,
+              ts,
+              freezeCount: asNumber(stat.freezeCount) ?? 0,
+            })
+
+            const kbps = d.ts > 0 ? Math.max(0, (d.bytes * 8) / d.ts) : 0
+            const lostTotal = d.packetsLost + d.packets
+            const lossPct =
+              lostTotal > 0
+                ? Math.max(0, Math.min(100, (d.packetsLost / lostTotal) * 100))
+                : 0
+
+            // Resolve codec + source track.
+            const codecStat = byId.get(asString(stat.codecId) ?? '')
+            let msTrackId = asString(stat.trackIdentifier)
+            if (!msTrackId && isUp) {
+              const mediaSource = byId.get(asString(stat.mediaSourceId) ?? '')
+              msTrackId = asString(mediaSource?.trackIdentifier)
+            }
+            const rid = asString(stat.rid)
+            const context = msTrackId ? contexts.get(msTrackId) : undefined
+            const baseLabel =
+              context?.label ??
+              `${kind ?? 'media'} ssrc ${asNumber(stat.ssrc) ?? '?'}`
+
+            const width = asNumber(stat.frameWidth)
+            const height = asNumber(stat.frameHeight)
+
+            tracks.push({
+              key,
+              dir: isUp ? 'up' : 'down',
+              kind: kind ?? 'unknown',
+              label: rid ? `${baseLabel} [${rid}]` : baseLabel,
+              codec: shortCodec(asString(codecStat?.mimeType)),
+              kbps,
+              fps: asNumber(stat.framesPerSecond),
+              res: width && height ? `${width}x${height}` : undefined,
+              lossPct,
+              limitation:
+                asString(stat.qualityLimitationReason) === 'none'
+                  ? undefined
+                  : asString(stat.qualityLimitationReason),
+              layer:
+                !isUp && kind === 'video' && context
+                  ? resolveLayerReason(context, height)
+                  : undefined,
+              froze: !isUp && kind === 'video' && d.freezeCount > 0,
+            })
+          }
+        })
+      }
+
+      // Stable order (direction, then label): sorting by bitrate would
+      // reshuffle rows on every tick as kbps fluctuates.
+      tracks.sort((a, b) =>
+        a.dir === b.dir
+          ? a.label.localeCompare(b.label)
+          : a.dir === 'up'
+            ? -1
+            : 1
+      )
+
+      const next: StatsSnapshot = {
+        ts: Date.now(),
+        upKbps: Math.round(upKbps),
+        downKbps: Math.round(downKbps),
+        rttMs,
+        jitterMs,
+        availableOutKbps,
+        route,
+        turnProtocols: Array.from(turnProtocols).sort(),
+        tracks,
+      }
+      setHistory((h) => [...h.slice(-59), next])
+    }
+
+    void collect()
+    const id = window.setInterval(() => void collect(), intervalMs)
+    return () => {
+      cancelled = true
+      window.clearInterval(id)
+    }
+  }, [room, enabled, intervalMs])
+
+  return { snapshot: history[history.length - 1], history }
+}

src/frontend/src/features/layout/components/OneToOneFocusLayout.tsx

@@ -1,119 +0,0 @@
-import { memo } from 'react'
-import type { TrackReferenceOrPlaceholder } from '@livekit/components-core'
-import { styled } from '@/styled-system/jsx'
-import { cva } from '@/styled-system/css'
-import { ParticipantTile } from '@/features/participantTile/components/ParticipantTile'
-import { getTrackKey } from '@/features/layout/utils/trackSelection'
-
-type OneToOneFocusLayoutProps = {
-  mainTrack?: TrackReferenceOrPlaceholder
-  thumbnailTrack?: TrackReferenceOrPlaceholder
-  disableTileControls?: boolean
-  /** Controls thumbnail dimensions – 'pip' for small PiP window, 'room' for the main viewport. */
-  context?: 'pip' | 'room'
-}
-
-/**
- * Focus layout for 1-to-1 calls: one main tile filling the area (letterboxed)
- * with an optional thumbnail overlay at the bottom-right.
- *
- * Shared between PiP and the main room – pass `disableTileControls` in PiP
- * where hover controls should be hidden.
- */
-export const OneToOneFocusLayout = memo(
-  ({
-    mainTrack,
-    thumbnailTrack,
-    disableTileControls,
-    context = 'room',
-  }: OneToOneFocusLayoutProps) => {
-    return (
-      <FocusContainer>
-        {mainTrack && (
-          <MainSlot>
-            <ParticipantTile
-              key={getTrackKey(mainTrack)}
-              trackRef={mainTrack}
-              disableTileControls={disableTileControls}
-            />
-          </MainSlot>
-        )}
-        {thumbnailTrack && (
-          <Thumbnail context={context}>
-            <ParticipantTile
-              key={getTrackKey(thumbnailTrack)}
-              trackRef={thumbnailTrack}
-              disableTileControls={disableTileControls}
-            />
-          </Thumbnail>
-        )}
-      </FocusContainer>
-    )
-  }
-)
-OneToOneFocusLayout.displayName = 'OneToOneFocusLayout'
-
-const FocusContainer = styled('div', {
-  base: {
-    position: 'relative',
-    width: '100%',
-    height: '100%',
-    borderRadius: '8px',
-    overflow: 'hidden',
-    backgroundColor: 'primaryDark.100',
-    boxSizing: 'border-box',
-  },
-})
-
-const MainSlot = styled('div', {
-  base: {
-    width: '100%',
-    height: '100%',
-    borderRadius: '8px',
-    overflow: 'hidden',
-    '& .lk-participant-tile': {
-      width: '100%',
-      height: '100%',
-    },
-    '& .lk-participant-media-video': {
-      objectFit: 'contain',
-    },
-  },
-})
-
-const Thumbnail = styled(
-  'div',
-  cva({
-    base: {
-      position: 'absolute',
-      right: '1.25rem',
-      bottom: '1.25rem',
-      aspectRatio: '16 / 9',
-      borderRadius: '8px',
-      overflow: 'hidden',
-      boxShadow: 'md',
-      zIndex: 2,
-      '& .lk-participant-tile': {
-        width: '100%',
-        height: '100%',
-      },
-    },
-    variants: {
-      context: {
-        pip: {
-          width: '42%',
-          maxWidth: '220px',
-          minWidth: '140px',
-        },
-        room: {
-          width: '20%',
-          maxWidth: '320px',
-          minWidth: '180px',
-        },
-      },
-    },
-    defaultVariants: {
-      context: 'room',
-    },
-  })
-)

src/frontend/src/features/layout/components/StageLayout.tsx

@@ -12,8 +12,7 @@ import {
 import { Track } from 'livekit-client'
 import { useSnapshot } from 'valtio'
 import { clearPinnedTrack, layoutStore, setPinnedTrack } from '@/stores/layout'
-import { useEffect, useMemo, useRef } from 'react'
-import { OneToOneFocusLayout } from '@/features/layout/components/OneToOneFocusLayout'
+import { useEffect, useRef } from 'react'
 
 export const StageLayout = () => {
   const lastAutoFocusedScreenShareTrack =
@@ -37,29 +36,6 @@ export const StageLayout = () => {
     (track) => !isEqualTrackRef(track, pinnedTrackRef)
   )
 
-  const cameraTracks = useMemo(
-    () => tracks.filter((t) => t.source === Track.Source.Camera),
-    [tracks]
-  )
-
-  const isOneToOne =
-    !pinnedTrackRef &&
-    screenShareTracks.length === 0 &&
-    cameraTracks.length <= 2
-
-  const oneToOneMainTrack = useMemo(() => {
-    if (!isOneToOne) return undefined
-    const remote = cameraTracks.find((t) => !t.participant?.isLocal)
-    const local = cameraTracks.find((t) => t.participant?.isLocal)
-    return remote ?? local
-  }, [isOneToOne, cameraTracks])
-
-  const oneToOneThumbnailTrack = useMemo(() => {
-    if (!isOneToOne) return undefined
-    const local = cameraTracks.find((t) => t.participant?.isLocal)
-    return oneToOneMainTrack === local ? undefined : local
-  }, [isOneToOne, cameraTracks, oneToOneMainTrack])
-
   /* eslint-disable react-hooks/exhaustive-deps */
   // Code duplicated from LiveKit; this warning will be addressed in the refactoring.
   useEffect(() => {
@@ -111,12 +87,7 @@ export const StageLayout = () => {
 
   return (
     <>
-      {isOneToOne ? (
-        <OneToOneFocusLayout
-          mainTrack={oneToOneMainTrack}
-          thumbnailTrack={oneToOneThumbnailTrack}
-        />
-      ) : !pinnedTrackRef ? (
+      {!pinnedTrackRef ? (
         <div className="lk-grid-layout-wrapper" style={{ height: 'auto' }}>
           <GridLayout tracks={tracks} style={{ padding: 0 }}>
             <ParticipantTile />

src/frontend/src/features/pip/components/layout/PipFocusLayout.tsx

@@ -0,0 +1,86 @@
+import { memo } from 'react'
+import type { TrackReferenceOrPlaceholder } from '@livekit/components-core'
+import { styled } from '@/styled-system/jsx'
+import { ParticipantTile } from '@/features/participantTile/components/ParticipantTile'
+import { getTrackKey } from '../../utils/pipTrackSelection'
+
+type PipFocusLayoutProps = {
+  mainTrack?: TrackReferenceOrPlaceholder
+  thumbnailTrack?: TrackReferenceOrPlaceholder
+}
+
+/**
+ * Focus layout used when 1-2 tracks are visible in the PiP window.
+ *
+ * The main tile is letterboxed (object-fit: contain) so the camera is
+ * never stretched to a non-video aspect and leaves dark padding
+ * above/below when the window shape doesn't match the source.
+ * The thumbnail keeps the usual cover fill.
+ */
+export const PipFocusLayout = memo(
+  ({ mainTrack, thumbnailTrack }: PipFocusLayoutProps) => {
+    return (
+      <FocusContainer>
+        {mainTrack && (
+          <MainSlot>
+            <ParticipantTile
+              key={getTrackKey(mainTrack)}
+              trackRef={mainTrack}
+              disableTileControls
+            />
+          </MainSlot>
+        )}
+        {thumbnailTrack && (
+          <Thumbnail>
+            <ParticipantTile
+              key={getTrackKey(thumbnailTrack)}
+              trackRef={thumbnailTrack}
+              disableTileControls
+            />
+          </Thumbnail>
+        )}
+      </FocusContainer>
+    )
+  }
+)
+PipFocusLayout.displayName = 'PipFocusLayout'
+
+const FocusContainer = styled('div', {
+  base: {
+    position: 'relative',
+    width: '100%',
+    height: '100%',
+    borderRadius: '8px',
+    overflow: 'hidden',
+    backgroundColor: 'primaryDark.100',
+    boxSizing: 'border-box',
+  },
+})
+
+const MainSlot = styled('div', {
+  base: {
+    width: '100%',
+    height: '100%',
+    borderRadius: '8px',
+    overflow: 'hidden',
+    '& .lk-participant-media-video': {
+      objectFit: 'contain',
+    },
+  },
+})
+
+const Thumbnail = styled('div', {
+  base: {
+    position: 'absolute',
+    right: '1.25rem',
+    bottom: '1.25rem',
+    width: '42%',
+    maxWidth: '220px',
+    minWidth: '140px',
+    aspectRatio: '16 / 9',
+    borderRadius: '8px',
+    overflow: 'hidden',
+    boxShadow: 'md',
+    zIndex: 2,
+  },
+})

src/frontend/src/features/pip/components/layout/PipGridLayout.tsx

@@ -4,7 +4,7 @@ import { styled } from '@/styled-system/jsx'
 import { ParticipantTile } from '@/features/participantTile/components/ParticipantTile'
 import { usePipElementSize } from '../../hooks/usePipElementSize'
 import { computePipGridLayout } from '../../utils/pipGrid'
-import { getTrackKey } from '@/features/layout/utils/trackSelection'
+import { getTrackKey } from '../../utils/pipTrackSelection'
 
 type PipGridLayoutProps = {
   tracks: TrackReferenceOrPlaceholder[]

src/frontend/src/features/pip/components/layout/PipScreenShareLayout.tsx

@@ -2,7 +2,7 @@ import { memo } from 'react'
 import type { TrackReferenceOrPlaceholder } from '@livekit/components-core'
 import { styled } from '@/styled-system/jsx'
 import { ParticipantTile } from '@/features/participantTile/components/ParticipantTile'
-import { getTrackKey } from '@/features/layout/utils/trackSelection'
+import { getTrackKey } from '../../utils/pipTrackSelection'
 
 type PipScreenShareLayoutProps = {
   screenShareTrack: TrackReferenceOrPlaceholder

src/frontend/src/features/pip/components/layout/PipStage.tsx

@@ -2,7 +2,7 @@ import React, { useMemo } from 'react'
 import { usePagination, useTracks } from '@livekit/components-react'
 import { RoomEvent, Track } from 'livekit-client'
 import { styled } from '@/styled-system/jsx'
-import { OneToOneFocusLayout } from '@/features/layout/components/OneToOneFocusLayout'
+import { PipFocusLayout } from './PipFocusLayout'
 import { PipGridLayout } from './PipGridLayout'
 import { PipPagination } from './PipPagination'
 import { PipScreenShareLayout } from './PipScreenShareLayout'
@@ -59,11 +59,9 @@ export const PipStage = () => {
     if (cameraTracks.length <= 1) {
       return (
         <StageFrame>
-          <OneToOneFocusLayout
+          <PipFocusLayout
             mainTrack={screenShareTrack}
             thumbnailTrack={cameraTracks[0]}
-            disableTileControls
-            context="pip"
           />
         </StageFrame>
       )
@@ -101,12 +99,7 @@ export const PipStage = () => {
 
   return (
     <StageFrame>
-      <OneToOneFocusLayout
-        mainTrack={mainTrack}
-        thumbnailTrack={thumbnailTrack}
-        disableTileControls
-        context="pip"
-      />
+      <PipFocusLayout mainTrack={mainTrack} thumbnailTrack={thumbnailTrack} />
     </StageFrame>
   )
 }

src/frontend/src/features/rooms/components/Conference.tsx

@@ -43,6 +43,7 @@ import { useSnapshot } from 'valtio'
 import { userPreferencesStore } from '@/stores/userPreferences'
 import { userStore } from '@/stores/user'
 import { WatchMediaDeviceErrors } from './WatchMediaDeviceErrors'
+import { MeetDevtools } from '@/features/devtools'
 import { VOICE_AUDIO_CONSTRAINTS } from '@/features/rooms/livekit/utils/constants'
 
 export const Conference = ({
@@ -298,6 +299,7 @@ export const Conference = ({
           <VideoConference />
           {!isMobile && <InviteDialog mode={mode} />}
           <PictureInPictureConference />
+          <MeetDevtools />
         </LiveKitRoom>
       </Screen>
     </QueryAware>

src/frontend/src/features/subtitle/component/Subtitles.tsx

@@ -75,15 +75,15 @@ const useTranscriptionState = () => {
     const segment = segments[0]
 
     setTranscriptionSegments((prevSegments) => {
-      const existingSegmentIds = new Set(prevSegments.map((s) => s.id))
-      if (existingSegmentIds.has(segment.id)) return prevSegments
-      return [
-        ...prevSegments,
-        {
-          participant: participant,
-          ...segment,
-        },
-      ]
+      const existingIndex = prevSegments.findIndex(
+        (s: TranscriptionSegmentWithParticipant) => s.id === segment.id
+      )
+      if (existingIndex === -1) {
+        return [...prevSegments, { participant, ...segment }]
+      }
+      const next = prevSegments.slice()
+      next[existingIndex] = { ...next[existingIndex], ...segment }
+      return next
     })
   }
 

src/helm/env.d/common.yaml.gotmpl

@@ -186,8 +186,6 @@ backend:
     CELERY_BROKER_URL: redis://default:pass@redis-master:6379/1
     # Recording & Transcription
     RECORDING_ENABLE: True
-    RECORDING_STORAGE_EVENT_ENABLE: True
-    RECORDING_STORAGE_EVENT_TOKEN: password
     SUMMARY_SERVICE_ENDPOINT: http://meet-summary:80/api/v2/async-jobs/transcribe/
     SUMMARY_SERVICE_API_TOKEN: password
     SUMMARY_SERVICE_WEBHOOK_API_TOKEN: webhook-password

src/helm/extra/templates/minio.yaml

@@ -113,26 +113,3 @@ spec:
             exit 0
       restartPolicy: Never
   backoffLimit: 3
----
-apiVersion: batch/v1
-kind: Job
-metadata:
-  name: minio-webhook
-spec:
-  template:
-    spec:
-      containers:
-      - name: mc
-        image: minio/mc
-        command:
-          - /bin/sh
-          - -c
-          - |
-            /usr/bin/mc alias set meet http://minio:9000 meet password && \
-            /usr/bin/mc admin config set meet notify_webhook:meet-webhook endpoint="https://meet.127.0.0.1.nip.io/api/v1.0/recordings/storage-hook/" auth_token="Bearer password" && \
-            /usr/bin/mc admin service restart meet --wait --json && \
-            sleep 15 && \
-            /usr/bin/mc event add meet/meet-media-storage arn:minio:sqs::meet-webhook:webhook --event put --prefix "recordings" && \
-            exit 0
-      restartPolicy: Never
-  backoffLimit: 3

src/summary/README.md

@@ -22,13 +22,6 @@ Configure your env values in `env.d/summary` to properly set up WhisperX and the
 make run
 ```
 
-When the stack is up, configure the MinIO webhook
-*(TODO: add this step to `make bootstrap`)*
-
-```sh
-make minio-webhook-setup
-```
-
 If you want to develop on the Celery workers with hot reloading, run:
 
 ```sh