Run the Example

Three example projects show a full working setup, all in this repository: a minimal bridge embedding (examples/basic-bridge) and two ready-to-run agents (examples/voice-agent and examples/avatar-agent). This page walks through them so you understand every moving part.

What a working setup needs

A LiveKit call has three processes, two of them yours:

1. Your agent worker - registers with your LiveKit project under an agent_name and waits for dispatch.
2. This bridge - creates a room per Teams call, dispatches the agent into it, relays audio.
3. StandIn (hosted) - joins the Teams call and dials the bridge.

1. Run the example voice agent

The example agents are plain LiveKit agents - nothing Teams-specific. The pipeline is picked from what your environment offers: Azure speech-to-speech realtime, Azure STT/LLM/TTS, or plain OpenAI (see the example’s README).

git clone https://github.com/komaa-com/livekit-msteams-bridge-py
cd livekit-msteams-bridge-py/examples/voice-agent
cp .env.example .env   # LIVEKIT_URL, LIVEKIT_API_KEY, LIVEKIT_API_SECRET + one model stack
uv sync
uv run python -m livekit.agents download-files
uv run worker.py dev

(Plain pip works too: pip install -r requirements.txt && python worker.py dev.)

The worker registers as standin-agent and waits - it will not join anything until the bridge creates a room and dispatches it.

2. Run the bridge example

pip install livekit-msteams-bridge
git clone https://github.com/komaa-com/livekit-msteams-bridge-py
cd livekit-msteams-bridge-py/examples/basic-bridge
cp .env.example .env   # same LiveKit project + LIVEKIT_AGENT_NAME=standin-agent + BRIDGE_SECRET
python main.py

It prints the WebSocket URL to give StandIn:

Point your StandIn identity's agent WebSocket URL at ws://<this-host>:9442/msteams/calling

The main.py is the recommended embedding shape in ~25 lines: load_dotenv(), load_config() (fails loud on any misconfiguration), await start_server(cfg), and a graceful await server.close() on Ctrl-C / SIGTERM that ends live calls with a spoken-protocol session.end rather than a hard drop.

3. Connect StandIn and call

1. Expose port 9442 with a tunnel (tailscale funnel --bg --set-path /msteams/calling http://127.0.0.1:9442/msteams/calling, which serves the bridge at wss://<your-tailnet-host>/msteams/calling with no port; cloudflared tunnel --url http://localhost:9442; or ngrok http 9442).
2. In your StandIn dashboard, set the identity’s Agent voice URL to the wss://.../msteams/calling form and make sure the shared secret equals BRIDGE_SECRET.
3. Call your Teams bot (or join the sandbox meeting). The bridge creates the room, dispatches standin-agent, and the agent answers.

4. Swap in the avatar agent

examples/avatar-agent is the same pipeline plus a lip-synced bitHuman avatar. Two extra variables in that agent’s .env (BITHUMAN_API_SECRET, BITHUMAN_MODEL_PATH), then:

cd ../avatar-agent && uv sync && uv run worker.py dev

Stop the voice worker first: both register as standin-agent, and explicit dispatch resolves a single name. The caller hears the avatar’s audio; the avatar’s video stays in the room (the Teams tile is rendered by StandIn’s own animated avatar - see Agents and Dispatch).

What the example agents demonstrate

Each example shows the three integration points your own agent can use:

• agent_name in WorkerOptions - the dispatch contract with LIVEKIT_AGENT_NAME.
• ctx.job.metadata - per-call caller context (caller_name, tenant_id, call_direction, user_id when known) for greetings and personalization.
• msteams.context / msteams.goodbye / msteams.vision data topics - call context, the governor’s goodbye handler (interrupt the current turn, speak the line), and the opt-in ambient-vision byte stream.

Details and copy-paste handlers: Agents and Dispatch.

From example to your own service

• Keep your own agent worker exactly as it is for WebRTC users - just give it an agent_name.
• Embed the bridge (await start_server(load_config())) or run the stock CLI.
• Set the governor variables (MAX_CALL_MINUTES, GOODBYE_TEXT) before production.
• For tests, inject a fake room with the connect_room argument - see Library API.