Oduist Connect Documentation

Pipecat AI Agent Setup

The Pipecat integration requires the connect_pipecat addon, a FreeSWITCH image containing mod_audio_fork, and the oduist/pipecat-agent:1.0.0 sidecar. Pipecat does not run inside Odoo.

1. Build the images

Build the FreeSWITCH image from connect_freeswitch/deploy (requires connect_freeswitch 19.0.2.1.0 or later). Use the image tag pinned in connect_freeswitch/deploy/docker-compose.yml and docker-compose.full.yml — oduist/freeswitch:2.1.2 at the time of writing:

docker build --platform linux/amd64 --provenance=false --sbom=false \
  -t oduist/freeswitch:2.1.2 connect_freeswitch/deploy

Build the sidecar:

docker build --platform linux/amd64 --provenance=false --sbom=false \
  -t oduist/pipecat-agent:1.0.0 connect_pipecat/deploy

connect_pipecat/deploy/oduflow-preset.yaml contains the service variables and WSS routing shape.

2. Pair Odoo and the sidecar

Open Connect → FreeSWITCH → Configuration → Settings → Pipecat AI.

  1. Set Pipecat Sidecar URL to its externally reachable base URL, for example wss://voice.example.com. Do not append /ws.
  2. Generate a URL-safe token of at least 24 characters and enter it as the Pipecat service token.
  3. Set exactly the same value as PIPECAT_SERVICE_TOKEN in the sidecar.
  4. Set ODOO_URL in the sidecar to Odoo's HTTPS base URL.
  5. Enter API keys for every provider used by an agent.
  6. Restart the sidecar and click CHECK STATUS. A healthy service reports UP (1.0.0).

The token authenticates both directions: FreeSWITCH uses Basic auth to open the media WebSocket; the sidecar uses Bearer auth for Odoo and /health.

3. Configure an agent

Go to Connect → FreeSWITCH → AI Agents, create an agent, and configure:

  • system prompt and optional greeting;
  • STT, LLM and TTS providers/models;
  • language, voice and maximum duration;
  • optional human transfer extension;
  • call recording if transcript audio should also be retained.

Use the Extension button to assign a unique extension number. A DID may then route to that extension using the normal Connect routing controls.

Verification

Call the extension and confirm the greeting is heard. Speak over a long agent reply: playback should stop immediately (killAudio) and the caller's new turn should be processed. Ask for a human and verify the configured extension rings. After hangup, open the call in Connect and verify summary and transcript.

Check these logs when diagnosing failures:

  • mod_audio_fork::connect_failed in FreeSWITCH: WSS, certificate or Basic token mismatch;
  • sidecar 401: Odoo/service token mismatch;
  • provider initialization error: missing key or unsupported model/voice;
  • call result 404: CDR creation was delayed beyond the sidecar retry window.

The acceptance target is first agent audio in under 1.4 seconds with working barge-in. Measure this in the deployment because provider and network latency dominate the result.