deepgram-python-voice-agent

Build a real-time voice agent in Python by using a WebSocket-based SDK blueprint that connects to Deepgram, managing full-duplex audio streaming, STT, TTS, and LLM interactions within a single session.

Function calling in a live voice agent session triggers tool executions from the model and handles responses directly within the same WebSocket connection, allowing real-time interaction without breaking the audio stream.

You can dynamically swap prompts, speak, or think settings mid-session without disconnecting, enabling seamless updates to the voice agent's behavior during a live conversation.

Manage conversation history and handle reconnects by using durable connections and context persistence, allowing the voice agent to resume conversations automatically after unexpected disconnects.

The Deepgram Python SDK supports full-duplex audio streaming with live conversation capabilities, allowing users to interrupt or barge-in during text-to-speech playback for natural interactions.

Maintaining a WebSocket connection to agent.deepgram.com requires correct provider configurations, robust error handling, and keep-alive management to ensure stable streaming audio and mid-session updates.