# Setup Guide

## 1. Install Python dependencies

```bash
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
```

## 2. Download F5-TTS model weights

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

## 3. Prepare your voice reference audio

F5-TTS clones voice from a reference clip. You need a clean 10-second WAV file of the voice you want to use.

1. Place it at `delamain_ref_10s.wav` in the project root
2. Edit `REF_TEXT` in `tts_engine.py` to match exactly what is spoken in the clip

For the authentic Cyberpunk 2077 Delamain voice, source a clean audio clip from the game and transcribe it.

## 4. Configure environment

```bash
cp .env.example .env
```

Edit `.env`:

| Variable | Required | Description |
|---|---|---|
| `LLM_URL` | Yes | OpenAI-compatible chat endpoint with tool calling |
| `LLM_MODEL` | Yes | Model id the endpoint serves (see its `/v1/models`) |
| `LLM_REASONING_EFFORT` | No | `none` turns thinking off on reasoning models |
| `SEARXNG_URL` | No | SearXNG instance for web search and the news fallback |
| `OBSIDIAN_VAULT_PATH` | No | Vault for notes and memory |
| `ALLOWED_ORIGINS` | Yes | Your domain/IP for CORS |

## 5. Set a password

```bash
python auth.py set-password yourpassword
```

Tokens expire after 30 days. To show the current token:

```bash
python auth.py show-token
```

## 6. Start services

**Backend:**
```bash
uvicorn main:app --host 0.0.0.0 --port 8888 --ws-ping-interval 0
```

**Frontend (development):**
```bash
cd frontend && npm install && npm run dev
```

**Frontend (production build):**
```bash
cd frontend && npm run build
# Serve dist/ with nginx or any static file server
```

## 7. Use it in the car

Open the frontend's address in the browser on your phone or on the Android Auto display, log in, and press the mic. Over the internet, put it behind HTTPS (for example a Cloudflare Tunnel) and add that origin to `ALLOWED_ORIGINS`. Browsers only allow the microphone and location on secure pages.