InitialDelaySeconds was set to 30s which caused the tilt stack
startup to be very slow. In this commit I split the initial delay
and the period arg.
5s seems to be relevant given the nature of the app.
With the introduction of file background uploads, only trigger the
webhook for files related to recordings.
Avoid firing the "recording saved" event for other file uploads,
preventing unnecessary queries and false triggers.
Migrate main meet app to use UV for dependancy management.
Also optimized the backend image build sequence for faster rebuilds
when dependencies don't change.
Also removed compiled django translations files are they are done in the build
process now.
Changes inspired by drive repo.
The wrong port was configured, causing CORS issues
when uploading files directly to the MinIO bucket.
Port 9001 is reserved for the MinIO console;
use the correct API port instead.
* Add a simple celery-backend deployment to helm chart, which
uses the same docker image as the backend, and runs a single
worker.
* Update dev keycloack values accordingly
* Update Tiltfile to properly set service dependencies
* Update minio deployment to increase the default max body size limit
(relevant to uploading files)
Add media_files_svc and ingress_media_files to helm chart
for serving medias related to the new files.
This setup is more straightforward than merging the
existing recording related media ones.
Also updated values to enable file upload by default in dev.
Mostly done by @lebaudantoine.
For the coming features we will need to store files on the meet side.
(for instance user backgrounds).
This commits adds a new Model to manage files, and the associated
serializers & viewsets. All are tested.
This work was heavily inspired by the work done by our friends at
https://github.com/suitenumerique/drive
It build on the same architecture design (upload directly to S3 but
download goes through our proxy), but model is much much simplier
(no folders, no file sharing, etc.).
Create a separate Ingress resource to isolate traffic targeting the
webhook-livekit endpoint and allow applying specific NGINX
annotations to this route.
Use an exact path match to take precedence over the default /api
regex rule defined in the base Ingress.
No similar change is made for the S3 webhook endpoint, as this
dependency will be removed from the project.
Configure the external application API across different Kubernetes setups
to enable seamless usage without repeated configuration
when iterating on endpoints.
Allow configuring whether a VAD model runs before calling an external ASR API.
Running VAD can save API calls (and costs) when no audible sound is detected,
but comes with the trade-off of additional computational overhead.
After upgrading Python to 3.13, not all development environments were
updated accordingly. This fixes the incorrect volume mount path
introduced by that upgrade.
Fix MinIO client configuration: I was incorrectly passing a full URL instead of
an endpoint, which caused errors in staging. Local development values did not
reflect the staging setup and were also out of sync with the backend.
Fix an unexpected behavior where filtering LiveKit webhook events sometimes
failed because the room name was not reliably extracted from the webhook data,
causing notifications to be ignored.
Configure the same filtering logic locally to avoid missing this kind of issue
in the future.
Pin egress to the production version, which uses a more recent release than the
default chart value (1.9.0).
Using the default could have led to issues; hopefully this change avoids them.
Link the transcription document to its related recording by adding a short
header explaining that users can download the audio file via a dedicated link.
This was a highly requested feature, as many users need to keep their audio
files.
As part of a small refactor, remove the argument length check in the metadata
analytics class. The hardcoded argument count made code evolution harder and was
easy to forget updating. Argument unwrapping remains fragile and should be
redesigned later to be more robust.
The backend is responsible for generating the download link to ensure
consistency and reliability.
I tried adding a divider, but the Markdown-to-Yjs conversion is very lossy and
almost never handles it correctly. Only about one out of ten conversions works
as expected.
Not all self-hosted instances will configure this setting, so a default text is
shown when the destination is unknown.
This is important to let users quickly click the link and understand which
platform is used to handle the transcription documents.
Explicitly explain that transcription is reserved for public servants. Remove
the temporary beta form: the feature is now available to all public servants,
with restrictions based on domain. Make white-labeling rules explicit and
clarify who to contact for access.
The beta form created frustration, with users registering and never hearing
back from the team.
Improve guidance when a user may be the meeting host but is not logged in, and
therefore cannot activate recording. Add a clear hint and a quick action to log
in. This decision is based on frequent support requests where users could not
understand why recording was unavailable while they were simply not logged in.
This release adds support for injecting custom Kubernetes
resources through the extraManifests parameter.
New features:
- Add extraManifests support for deploying custom resources
- Support multiple input formats (list, map, raw YAML strings)
- Enable Helm template variables in injected manifests
Add ability to inject custom Kubernetes manifests through the
values.yaml file. This allows users to deploy additional
resources (Deployments, Services, ConfigMaps, etc.) without
modifying the chart templates.
The template supports multiple input formats: list of objects,
map of named objects, and raw YAML strings, providing maximum
flexibility for users.
- Create templates/extra-objects.yaml with flexible rendering
- Add extraManifests parameter in values.yaml with documentation
- Support Helm template variables in injected manifests
- Handle list, map, and string YAML formats automatically
Add dynamic configuration for Deepgram STT via environment variables,
enabling multilingual real-time subtitles with automatic language
detection.
Changes:
- Add DEEPGRAM_STT_* environment variables pattern for configuration
- Implement _build_deepgram_stt_kwargs() to dynamically build STT
parameters from environment variables
- Add whitelist of supported parameters (model, language) for LiveKit
Deepgram plugin
- Log warnings for unsupported parameters (diarize, smart_format, etc)
- Set default configuration: model=nova-3, language=multi
- Document supported parameters in Helm values.yaml
Configuration:
- DEEPGRAM_STT_MODEL: Deepgram model (default: nova-3)
- DEEPGRAM_STT_LANGUAGE: Language or 'multi' for automatic detection
of 10 languages (en, es, fr, de, hi, ru, pt, ja, it, nl)
Note: Advanced features like diarization and smart_format are not
supported by the LiveKit Deepgram plugin in streaming mode.
Introduce pop-in alerting participants of automatic 2-minute idle
disconnect to enable LiveKit node configuration updates during
maintenance windows, preventing forgotten tabs from blocking
overnight production updates following patterns
from proprietary videoconference solutions.
Without explicit commands in values.yaml,
celeryTranscribe and celerySummarize pods
were using the Dockerfile's default CMD (uvicorn),
which started the REST API instead of Celery workers.
This fix adds default commands to values.yaml for both services,
ensuring they run as Celery workers processing their respective
queues (transcribe-queue and summarize-queue).
This chart exposes an external API from the backend pod.
Currently, it does not include conditional addition of the external API route.
This functionality will be added later.
Make WhisperX language detection configurable through FastAPI settings
to handle empty audio start scenarios where automatic detection fails and
incorrectly defaults to English despite 99% French usage.
Quick fix acknowledging long-term solution should allow dynamic
per-recording language selection configured by users through web
interface rather than global server settings.
Prepare for the introduction of new endpoints reserved for external
applications. Configure the required router and update the Helm chart to ensure
that the Kubernetes ingress properly routes traffic to these new endpoints.
It is important to support independent versioning of both APIs.
Base route’s name aligns with PR #195 on lasuite/drive, opened by @lunika
Remove incorrect whitespace in queue names that prevented Celery
workers from listening to proper queues. Workers were attempting to
connect to non-existent queues, breaking task distribution.
Implement automated MinIO webhook configuration using Kubernetes job
to enable recording feature functionality. This eliminates manual
setup requirements and ensures consistent webhook configuration
across deployments.
Restore certificate mounting for MinIO webhook communication to
backend after migrating away from unmaintained Bitnami chart.
Mount certificate in proper volume to enable secure bucket-to-backend
webhook delivery.
Add Celery summarize and transcribe worker configuration to Helm
charts for summary microservice. Create new deployment resources
and increment chart version to support distributed task processing.
Implement summarization functionality that processes completed meeting
transcripts to generate concise summaries.
First draft base on a simple recursive agentic scenario.
Observability and evaluation will be added in the next PRs.
Rename incorrectly named OpenAI configuration settings since
they're used to instantiate WhisperX client which is not OpenAI
compatible, preventing confusion about actual service dependencies.
Add required whitespace between braces in template directives to
comply with Kubernetes rule S6893.
Improves template readability and follows Kubernetes best practices for
template formatting and maintainability.
Allow any user, anonymous or authenticated, to start subtitling
in a room only if they are an active participant of it.
Subtitling a room consists of starting the multi-user transcriber agent.
This agent forwards all participants' audio to an STT server and returns
transcription segments for any active voice to the room.
User roles in the backend room system cannot be used
to determine subtitle permissions.
The transcriber agent can be triggered multiple times but will only join a
room once. Unicity is managed by the agent itself.
Any user with a valid LiveKit token can initiate subtitles. Feature flag
logic is implemented on the frontend. The frontend ensures the "start
subtitle" action is only available to users who should see it. The backend
does not enforce feature flags in this version.
Authentication in our system does not imply access to a room. The only
valid proof of access is the LiveKit API token issued by the backend.
Security consideration: A LiveKit API token is valid for 6 hours and
cannot be revoked at the end of a meeting. It is important to verify
that the token was issued for the correct room.
Calls to the agent dispatch endpoint must be server-initiated. The backend
proxies these calls, as clients cannot securely contact the agent dispatch
endpoint directly (per LiveKit documentation).
Room ID is passed as a query parameter. There is currently no validation
ensuring that the room exists prior to agent dispatch.
TODO: implement validation or error handling for non-existent rooms.
The backend does not forward LiveKit tokens to the agent. Default API
rate limiting is applied to prevent abuse.
Create basic Helm chart for LiveKit agent framework deployment on
Kubernetes, inspired by meet-summary FastAPI server configuration.
Integrate chart into local tilt development stack and properly handle
certificate issues that typically occur when calling LiveKit server
with nip.io domain names.
Remove call to generate demo data in tilt stack as it was never useful
to developers and only complicated the migration job unnecessarily.
Migration job should be laser focused on applying database migrations
rather than seeding mock data, improving clarity and reducing
complexity.
Replace mock Django secret key with longer version to resolve security
warnings in development stack.
Still not production-suitable as key remains versioned in repository,
but eliminates security warnings during development workflow.
Remove dependencies on bitnami Helm charts since recent changes in
bitnami organization led to charts no longer being maintained or
published.
Enhanced the Tilt dependencies to avoid any bootstrap or refresh
errors while developping using the Tilt stack.
Making components dependant from each others increase slightly
the time required to spin up the stack the first time.
Add default telephony configuration to the tilt stack to enable
development workflow around authentication features.
Note: This is a fake/mock configuration and is not functional for
production use. It's intended solely for development purposes.
The LiveKit API URL is necessary to interact with the API. It uses https
protocol.
Eplicit wss protocol is necessary in Websocket constructor for some
older browsers.
This resolves critical compatibility issues with legacy browsers
(notably Firefox <124, Chrome <125, Edge <125) that lack support
for HTTPS URLs in the WebSocket() constructor. Without explicit WSS
URLs, WebSocket signaling connections may fail, crash, or be blocked
entirely in these environments.
The setting is optional and defaults to the current behavior when
not specified, ensuring zero breaking changes for existing deployments.
Connection warmup wasn't working properly - only works when trying to
establish WebSocket first, then workaround kicks in. Call WebSocket
endpoint without auth info expecting 401 error, but enough to initiate
cache for subsequent WebSocket functionality.
Scope this **dirty** trick to Firefox users only. Haven't figured out
how to detect proxy from JS code simply.
Tested in staging and works on our constrained WiFi.
Add product analytics to understand summary feature usage and behavior.
Track transcript and task metadata for insights without exposing sensitive
content or speaker data.
Hacky but functional PostHog usage - fully optional for self-hosting.
Extensive tracking approach works for current needs despite not being
PostHog's typical use case.
Florent Chehab