​
Build and run with Docker

The repo ships a Dockerfile and helper scripts:
yarn docker:run exists as a shortcut, but it does not pass an env file. Use the explicit docker run above so the container receives your configuration.

​
Required configuration

Speech and telephony are optional and only needed for voice agents: LIVEKIT_URL, LIVEKIT_API_KEY, LIVEKIT_API_SECRET, LIVEKIT_SIP_HOST, VOICE_WEBHOOK_URL, plus an STT provider key such as DEEPGRAM_API_KEY or ASSEMBLYAI_API_KEY. Integration credentials follow a consistent naming pattern — {SERVICE}_CLIENT_ID, {SERVICE}_CLIENT_SECRET, {SERVICE}_REDIRECT_URI. See each integration guide for the exact names.

​
Database migration

Model schemas are applied with:
This runs a build first. Run it once on a fresh database, and again after upgrading to a release that changes models.

​
Running behind a load balancer

GoConva runs multi-instance in production behind a load balancer. Socket.IO emits do not cross instances — a real-time event emitted on instance A is not delivered to a client connected to instance B. If you are running more than one replica and relying on live updates, put a shared adapter (such as Redis) in front of Socket.IO, or pin sessions to an instance.
CORS preflight requests bypass the LoopBack sequence entirely — only the cors.origin callback runs. If you are debugging a preflight that fails, instrument that callback rather than the sequence.

​
The voice agent worker

Voice agents run as a separate Python process:
This invokes uv run python agent_workflow.py dev. It needs the LiveKit variables above and an STT provider key. Deploy it alongside the API, not inside the same container.

​
Health checks

Point your orchestrator’s liveness probe at: