Raspi-Agent is a modular streaming voice assistant for Raspberry Pi with built-in wake-word detection, streaming audio, and LLM back-end support (on-device or off-device)
- Wake word detection
- Streaming audio playback
- Natural conversation via OpenAI APIs
- Device registration and certificate enrollment
- Device-to-backend communication over mTLS
- RAG integration
- Agent memory
- Web dashboard
- Wake Word Detection — powered by Porcupine
- Streaming Audio Playback — real-time PCM or MP3 output via PortAudio
- Natural Conversation — integrates with OpenAI (STT, LLM, TTS)
- Dual Architecture — choose between:
- Onboard mode — runs all AI calls directly from the Pi
- Offboard mode — sends recordings to a backend for processing
Wake Word → Recorder → Voice Assistant → Player → User
Two orchestrator implementations:
| Mode | Description | Example |
|---|---|---|
| Onboard | Runs everything locally via OpenAI APIs (STT, LLM, TTS) | cmd/raspi-agent-onboard-local |
| Offboard | Streams recorded audio to a backend that processes it | cmd/raspi-agent-onboard |