https://github.com/ownerofglory/raspi-agent

DIY on-device voice assistant for Raspberry Pi: wake word, speech-to-text, LLM reasoning, and text-to-speech streaming via Bluetooth audio
https://github.com/ownerofglory/raspi-agent

agent ai diy go home-automation openai porcupine portaudio rag raspberry-pi self-hosted streaming voice-assistant

Last synced: about 2 months ago
JSON representation

DIY on-device voice assistant for Raspberry Pi: wake word, speech-to-text, LLM reasoning, and text-to-speech streaming via Bluetooth audio

• Host: GitHub
• URL: https://github.com/ownerofglory/raspi-agent
• Owner: ownerofglory
• License: mit
• Created: 2025-10-06T19:56:40.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-11-15T21:15:43.000Z (8 months ago)
• Last Synced: 2025-11-15T23:21:58.543Z (8 months ago)
• Topics: agent, ai, diy, go, home-automation, openai, porcupine, portaudio, rag, raspberry-pi, self-hosted, streaming, voice-assistant
• Language: Go
• Homepage:
• Size: 994 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Raspi agent (DIY voice assistance)

**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)

![](https://github.com/ownerofglory/raspi-agent/actions/workflows/build.yaml/badge.svg)

[![Quality Gate Status](https://sonarcloud.io/api/project_badges/measure?project=ownerofglory_raspi-agent&metric=alert_status)](https://sonarcloud.io/summary/new_code?id=ownerofglory_raspi-agent)

### *Currently, in development ...*

## Planned Features

- [x] Wake word detection

- [x] Streaming audio playback

- [x] Natural conversation via OpenAI APIs

- [x] Device registration and certificate enrollment

- [ ] Device-to-backend communication over mTLS

- [ ] RAG integration

- [ ] Agent memory

- [ ] Web dashboard



---

##  Features

- **Wake Word Detection** — powered by [Porcupine](https://picovoice.ai/platform/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

## Architecture Overview

> 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` |

---