A DIY voice-controlled smart display for your living room — on a Raspberry Pi 4.
Local-first · private · open source. A self-hosted alternative to an Echo Show or Nest Hub.
-02569B.svg?logo=flutter&logoColor=white){kind=icon}
PiBoard turns a Raspberry Pi 4 + an HDMI screen into a beautiful voice + touch smart display for the living room — meant to be seen from the couch, 2–3 m away.
Say the wake word ("Terminator, mets de la musique") or just tap. It plays music, shows the weather, plays YouTube, streams your cameras, drives your speakers and your smart home — and the entire voice pipeline can run 100% locally and for free: no cloud, no subscription, nothing leaving your home. Cloud services are an optional fallback, never a requirement.
It's a real, daily-driver project running in a living room in Guadeloupe. The goal of this repo is to let you run your own.
Why "PiBoard"? It's the homemade answer to the Echo Show / Nest Hub: you own the hardware, you own the data, and you can change literally anything — it's all here.
 Weather — hourly + 3-day, crisp SVG icons |
 Speakers — volume, EQ, night mode |
 Home — shutters, gate, smart plugs |
 Music — radio, library, Spotify, Deezer |
- 🗣️ Voice assistant — wake word → speech-to-text → intent/LLM → action → text-to-speech. A fast keyword router handles common commands with zero LLM calls; the LLM only answers open questions.
- 🎵 Music — internet radio (free, no account, the default), a local library, Spotify, or Deezer (opt-in). One clean provider interface.
- 🌤️ Weather — Open-Meteo (no API key), current + hourly + 3-day, with beautiful Meteocons SVG icons.
- 📺 YouTube — search and play, hardware-decoded in-app (no Chromium).
- 📷 Cameras — live snapshots & MJPEG streams (UniFi Protect).
- 🔊 Speakers — Devialet Phantom control over IP (volume, EQ, night mode) + AirPlay output.
- 🏠 Home automation — Shelly (Gen 1/2/3) & TP-Link Kasa out of the box, or plug into your existing Home Assistant. Devices live in config — no code changes.
- 🎨 Native, themeable, bilingual — a native flutter-pi UI (no web browser), 🇫🇷/🇬🇧 FR/EN, rebrandable colors without recompiling, and resolution-independent (runs on any landscape screen).
- 🔒 Local-first & private — the whole voice loop can run on your own hardware; cloud is optional.
- ⚙️ Web admin panel — configure everything from your phone/browser.
wake word ──▶ speech-to-text ──▶ intent / LLM ──▶ action ──▶ text-to-speech ──▶ speakers
Every stage is swappable, local or cloud:
| Stage | Local (free) | Cloud (optional fallback) |
|---|---|---|
| Wake word | livekit-wakeword (trained model) | — |
| Speech-to-text | Vosk (offline) · Nemotron ASR (LAN) | Voxtral (Mistral) |
| LLM | Ollama (e.g. Gemma) via a tiny LAN gateway | Mistral |
| Text-to-speech | Piper (offline) · Voxtral MLX (LAN) | Voxtral (Mistral) |
Pick all-local (zero cost, fully private), Pi + a LAN helper machine (a Mac/PC running Ollama),
or cloud — it's a couple of .env switches.
┌──────────────────────── Raspberry Pi 4 ────────────────────────┐
│ flutter-pi UI ⇄ WebSocket ⇄ FastAPI backend (Python) │
│ (native, HDMI) │ │
│ mic ▶ wake ▶ STT ▶ intent ▶ action ▶ TTS │
│ │ music · weather · video │
│ │ cameras · speakers · home │
└──────────────────────────┬─────────────────────────┬───────────┘
optional LAN AI helper AirPlay / speakers
(Ollama + TTS over HTTP)
- UI —
frontend-flutter/(native flutter-pi, no Chromium). - Backend —
backend/(FastAPI/asyncio); every service is zero-crash (degrade, never take down the loop). - Admin web —
frontend/admin/(Svelte). - LAN AI gateway —
lan-voice-gateway/(optional; wraps Ollama + a local TTS so the Pi stays light).
| Part | Notes |
|---|---|
| Raspberry Pi 4 (4 GB+) | Bookworm 64-bit |
| HDMI display | designed for 1920×1200 landscape; scales to any landscape resolution |
| USB microphone | a far-field array (e.g. ReSpeaker) works best |
| Speakers | anything PipeWire can reach — AirPlay, HDMI, Bluetooth, jack… (Devialet Phantom gets extra IP control) |
| (optional) a LAN machine | any PC/Mac to run Ollama + local TTS for the free LLM/voice path |
git clone https://github.com/elkir0/PiBoard.git
cd PiBoard
# 1) Configure (everything is optional — empty = feature disabled, cleanly)
cp .env.example .env # set your weather location, integrations, etc.
# 2) Backend
cd backend
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python main.py # serves the API + admin on :8000
# 3) UI (build the flutter-pi bundle on a dev machine, deploy to the Pi)
# see scripts/deploy-flutter-v3.shThen open the admin panel at http://<pi>:8000/admin/ (a random admin password is generated and
printed in the log on first run). The minimum to get going is a weather location and a reachable
audio output — everything else is progressive.
📜 Full configuration reference:
.env.example. 🗺️ Where the project is heading: see the Roadmap below.
Flutter (flutter-pi, native ARM, no browser) · Python (FastAPI + asyncio) · Svelte (admin) · PipeWire audio · Vosk / Piper / Ollama / Voxtral for the local voice stack.
Deezer support is opt-in and uses a personal ARL token (a grey area of Deezer's ToS — personal use, no stream archiving). No credentials ship in this repo; you provide your own. The legal default is free internet radio. Use it responsibly.
Shipped recently: legal-by-default music, full FR/EN localization, the first-run setup wizard, runtime theming, a pluggable home-automation layer (Shelly/Kasa or Home Assistant), resolution-independent landscape layouts, and a Spotify Connect receiver via go-librespot.
Next up:
- 📱 Portrait & adaptive layouts — full reflow for portrait screens and an adaptive nav rail.
- 🏠 Deeper Home Assistant integration — validate against a real HA instance and make the Home page fully data-driven.
- 🧰 Install/docs polish — hardware matrix, known-good shopping list, and clearer all-on-Pi / LAN-helper / cloud presets.
- 🗣️ More wake words & languages — bring-your-own hotword, additional UI locales.
Issues and PRs welcome. The project values simplicity over cleverness — each component as small and readable as possible, every service fails soft. Pick anything from the roadmap, or scratch your own itch.
MIT — do what you like, no warranty.
Built and maintained with AI-assisted development. Made for a real living room. 🛋️