Microsoft Teams Bridge for ElevenLabs Agents (Python)

Bridge Microsoft Teams voice/video calls to an ElevenLabs Agent.

PyPI package: elevenlabs-msteams-bridge - the -py suffix is only in this repository's name, to distinguish it from the Node.js sibling repo.

This is the Python sibling of @komaa/elevenlabs-msteams-bridge (Node.js) - same wire contract, same environment variables, drop-in interchangeable behind the same .env file. It terminates the StandIn media bridge wire protocol on one side and the ElevenLabs Agent WebSocket on the other:

• No transcoding: both sides speak base64 PCM 16 kHz mono - the hot path is copy-only.
• Barge-in: ElevenLabs interruptions map to playback flushes, with ghost-audio filtering.
• On-demand vision: the agent's look client tool answers from the caller's camera or screen-share, via your own OpenAI-compatible vision endpoint (frames never persisted) or via ElevenLabs multimodal upload (gated on Teams recording).
• Call governors: a bridge-side hard time cap with a deterministic TTS goodbye, plus the worker-side governor.
• Hardened: HMAC-signed upgrades with replay guard, connection caps, SSRF-guarded image fetches, dead-peer detection, graceful SIGTERM drain, Prometheus /metrics.

StandIn is the hosted media bridge that joins the Teams call and dials this bridge - you run no Teams media stack yourself.

Documentation: komaa-com.github.io/elevenlabs-msteams-bridge-py (getting started, example walkthrough, configuration and library reference, wire protocol). Teams/StandIn setup lives at docs.komaa.com.

Install

pip install elevenlabs-msteams-bridge

Requires Python 3.10+.

Run

ELEVENLABS_API_KEY=sk_... \
ELEVENLABS_AGENT_ID=agent_... \
WORKER_SHARED_SECRET=... \
elevenlabs-msteams-bridge

A .env file in the working directory is loaded automatically (existing environment wins). The bridge listens on ws://0.0.0.0:8080/voice/msteams/stream by default; StandIn appends /{callId} per call. Expose the port with a tunnel and register the wss:// URL as your identity's Agent voice URL in the StandIn dashboard.

Your ElevenLabs agent's audio input and output format must be PCM 16000 Hz - the bridge ends the call with a clear error if the agent negotiates anything else.

Embed

import asyncio
from elevenlabs_msteams_bridge import load_config, start_server

async def main():
    server = await start_server(load_config())
    await asyncio.Event().wait()  # run until cancelled

asyncio.run(main())

Pass your own async vision callable to answer the agent's look tool with any model you like - the raw frame never leaves your process:

async def describe(frame: dict, question: str) -> str:
    ...  # call your vision model with frame["dataBase64"] / frame["mime"]
    return "a person holding a badge"

server = await start_server(load_config(), vision=describe)

Configuration

Everything is environment variables; names are identical to the Node package.

Endpoints

• GET /healthz - liveness.
• GET /metrics - Prometheus counters (calls, rejections, relayed/dropped frames).
• GET /{...}/{callId} + WebSocket upgrade - the worker wire, HMAC-signed with X-OpenClawTeamsBridge-Timestamp / X-OpenClawTeamsBridge-Signature over "{timestampMs}.{callId}".

Notes for operators:

• /healthz and /metrics are unauthenticated (only the WebSocket upgrade is HMAC-gated). They expose no call content - just liveness and counters - but if you would rather not leak call volumes, keep the port behind your ingress/tunnel rules.
• One bridge process serves one agent id (ELEVENLABS_AGENT_ID). Run one process per agent if you route multiple agents.

Vision and recording

The look tool prefers your VISION_API_URL endpoint: the frame is described transiently and only the text enters the conversation. Without one, the bridge falls back to uploading the frame to ElevenLabs (multimodal) - that persists the frame with a third party, so it is only allowed while Teams recording is active. Note that even path-2 descriptions become ElevenLabs conversation content, which ElevenLabs retains per your agent's settings; enable the agent's zero-retention mode if callers' surroundings must not be stored.

License

MIT (c) Alaaeldin Elhenawy