Delamain

"Delamain online. I trust your day has been satisfactory so far."

Warning

Delamain is a hobby project made for fun. It is not a safety system. Do not interact with any interface, voice or otherwise, in a way that distracts you from driving. The authors accept no liability for accidents, injuries, traffic violations, or any other incidents arising from the use of this software. Use at your own risk. Always prioritise road safety above all else.

An AI companion for your vehicle — modeled on Delamain from Cyberpunk 2077. Delamain integrates with sunnypilot on Comma hardware to deliver real-time voice commentary, proactive driving observations, navigation, web search, and a full data dashboard. Access it from your phone or via Android Auto's web browser — no app installation required.

---

Screenshots

---

Features

• Voice personality — Delamain speaks with the exact character from Cyberpunk 2077: formal, measured, dry wit. Powered by F5-TTS with a custom voice clone.
• Real-time vehicle events — hard braking, ACC engagement, lead car proximity, speeding alerts, lane changes, thermal warnings, and more — all with Delamain-voiced responses.
• Static phrase pre-cache — alert phrases are synthesized at startup and served instantly. No latency on common events.
• Proactive commentary — Delamain speaks unprompted after 30 minutes of driver silence while moving.
• Navigation — ask Delamain to navigate anywhere; geocoded via Nominatim.
• Web search — live weather, traffic, news, fuel prices via SearXNG integration.
• Road camera vision — ask what's ahead; the sunnypilot bridge captures a road frame and Delamain describes it.
• Data dashboard — web GUI with telemetry charts, event log, conversation history, trip summaries with GPS route maps and drive scores.
• JWT authentication — single shared password, rate-limited login, 30-day tokens.

---

Architecture

┌─────────────────────┐     WebSocket      ┌──────────────────────┐
│  Comma 4            │◄──────────────────►│  Delamain Backend    │
│  sunnypilot bridge  │  vehicle telemetry  │  (FastAPI + F5-TTS)  │
└─────────────────────┘  events / snapshots └──────────┬───────────┘
                                                        │ REST / WS
                                            ┌───────────▼──────────┐
                                            │  Web UI              │
                                            │  (React + Vite)      │
                                            └──────────────────────┘

• Backend — FastAPI on Python 3.12, SQLite database, F5-TTS voice engine, llama.cpp LLM integration.
• Frontend — React + Vite, Tailwind CSS (cyber theme), recharts, Leaflet maps.
• Bridge — delamaind.py runs on the Comma device, reads sunnypilot msgq shared memory, streams telemetry and events over WebSocket.

---

Requirements

Hardware

• Comma 4 (or Comma 3X) running sunnypilot
• Server with NVIDIA GPU (RTX 3060+ recommended for F5-TTS inference; 8GB+ VRAM)
  • CPU-only works but TTS will be significantly slower on uncached phrases

Software

• Python 3.11+
• CUDA 12+ (if using GPU)
• Node.js 20+ (for frontend dev)
• llama.cpp server or any OpenAI-compatible LLM endpoint
• SearXNG (optional, for web search)

---

Installation

1. Clone and install dependencies

git clone https://github.com/vonhex/delamain.git
cd delamain
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Download F5-TTS model weights

mkdir -p checkpoints/F5TTS_Base
huggingface-cli download SWivid/F5-TTS \
  F5TTS_Base/model_1200000.pt \
  F5TTS_Base/vocab.txt \
  --local-dir checkpoints

3. Prepare your reference audio

F5-TTS clones the voice from a reference clip. Provide a clean 10-second WAV of the voice you want to clone:

cp your_voice_reference.wav delamain_ref_10s.wav

Edit REF_TEXT in tts_engine.py to match exactly what is spoken in that clip.

4. Configure environment

cp .env.example .env
# Edit .env with your LLM URL, SearXNG URL, and allowed origins

5. Set a password

python auth.py set-password <your-password>

6. Start the backend

uvicorn main:app --host 0.0.0.0 --port 8888 --ws-ping-interval 0

7. Start the frontend

cd frontend
npm install
npm run dev

Visit http://localhost:5173 and log in with your password.

---

Docker

cp .env.example .env   # edit .env first
docker compose up -d

Pre-built image: ghcr.io/vonhex/delamain:latest

F5-TTS model weights and reference audio are not included in the image. Mount them via volumes as shown in docker-compose.yml.

---

sunnypilot Bridge

Delamain connects to sunnypilot via a bridge daemon that runs on the Comma device.

→ vonhex/delamain-sp-bridge

---

Environment Variables

Variable	Default	Description
LLM_URL	http://localhost:8080/v1/chat/completions	OpenAI-compatible LLM endpoint
SEARXNG_URL	http://localhost:8887/search	SearXNG instance for web search
ALLOWED_ORIGINS	http://localhost:5173,http://localhost:4173	CORS allowed origins (comma-separated)

---

API

Endpoint	Auth	Description
POST /api/login	—	Authenticate, receive JWT
GET /health	—	Health check
WS /ws/{client_id}?token=	JWT	Main WebSocket — chat, events, audio
GET /api/data/telemetry	JWT	Vehicle telemetry history
GET /api/data/events	JWT	Vehicle event log
GET /api/data/conversations	JWT	Chat history
GET /api/data/event-counts	JWT	Event frequency
GET /api/data/trips	JWT	Trip summaries with GPS routes and drive scores
POST /api/chat	JWT	Single-turn chat (REST fallback)

---

License

PolyForm Noncommercial License 1.0.0 — free for personal use; commercial use requires a separate agreement.

---

Acknowledgements

• F5-TTS — voice synthesis
• sunnypilot — the ADAS platform Delamain bridges
• CD Projekt Red — Cyberpunk 2077 and the Delamain character that inspired this project