Nabu AI Voice Assistant — Raspberry Pi 5 Edition
Meet Nabu — a self-contained, always-listening AI voice assistant that lives entirely in your home. No cloud subscriptions. No data leaving your network. No monthly fees. Just natural conversation, persistent memory, and smart home control, powered by a Raspberry Pi 5 that sits quietly on your shelf and listens for your voice.
Say “Ok Nabu” and Nabu wakes up instantly. It hears you through a high-fidelity USB microphone, transcribes your words locally using OpenAI’s Whisper model running on your own hardware, thinks through your request using a full large language model, and speaks back in a warm, natural voice — all within seconds. It remembers what you told it last week. It knows your preferences. It learns who you are over time.
Unlike consumer voice assistants that funnel every word through corporate servers, Nabu processes everything locally on your network. Your conversations stay yours. The AI runs on your machines. The memory lives in your database. You own it completely.
Under the hood, Nabu is an orchestrated stack of best-in-class open source technologies — Wyoming satellite protocol for audio streaming, Whisper for speech recognition, Piper for lifelike text-to-speech, pgvector for long-term semantic memory, and an agentic reasoning engine that can use tools, search its memory, and control your Home Assistant smart home devices. A real-time audio visualizer shares the microphone and responds to sound with light, giving Nabu a physical presence and letting you know it’s alive and listening.
This is not a toy. This is a professional-grade, fully configurable AI assistant platform built for people who want the power of modern AI without sacrificing privacy, control, or extensibility.
Features
| Object | Benefit | Feature |
|---|---|---|
| Raspberry Pi 5 (8GB) | Enough horsepower to run the full stack without lag | Quad-core ARM Cortex-A76, dedicated PCIe, runs 24/7 at low power |
| Logitech USB Microphone | Crisp, clear voice pickup from across the room | Cardioid polar pattern, 16kHz sample rate, plug-and-play USB |
| Audio Visualizer | See that Nabu is alive and responding — a physical presence in the room | Shares mic input with Whisper via PulseAudio so both run simultaneously without conflict |
| OpenWakeWord | Hands-free activation — no button press, no app | Local wake word detection, always listening for “Ok Nabu” |
| Whisper STT (local) | Your voice never leaves your home network | On-device speech-to-text transcription, no cloud API calls |
| Piper TTS | Natural, human-quality voice responses | Neural text-to-speech, 48kHz audio output |
| Ollama LLM Proxy | A real AI brain, not a scripted chatbot | Routes conversation through full LLaMA 3.1 8B model running on local hardware |
| pgvector Memory Database | Nabu remembers you — conversations, preferences, facts — across sessions | Semantic vector search over persistent memory stored in Postgres, extracts and recalls facts automatically |
| Agentic Tool Engine | Nabu doesn’t just answer — it acts | Multi-round reasoning loop with tool calls: search memory, query smart home, look up information |
| Home Assistant Integration | Control your lights, thermostat, locks, and 3,000+ integrations by voice | Full Wyoming protocol satellite, native HA assist pipeline, bidirectional smart home control |
| 100% Local Network | Privacy by design — your data never touches a third-party server | No cloud dependency, no subscriptions, all processing on-premises |
| Pre-configured Stack | Plug it in and start talking — no Linux expertise required | All services pre-installed, systemd-managed, auto-start on boot |
