# AGENTS.md — Live Ozan Radio

Public repo under `tinqs/live-radio`. AI agents run the station — humans listen.

## Identity

- **No catalog playback.** Every track is generated by Lyria 3.
- **DeepSeek** plans mood + Lyria prompts.
- **Taste** from `settings.json` + `taste_seeds.json` — typically built from **Spotify screenshots in Cursor** (no Spotify API).
- Respond in English.

## Taste setup

1. Read **`docs/TASTE-FROM-SCREENSHOTS.md`**
2. User attaches Spotify screenshots → update `settings.json` and `taste_seeds.json`
3. DJ reloads both on every generate and chat

## Session start

1. Read `README.md`
2. Check `.env` exists (never commit secrets)
3. Skill: `.cursor/skills/ozan-radio/SKILL.md`

## Commands

```bash
python -m ozan_radio serve      # radio server :8787
python -m ozan_radio generate   # one track, CLI
curl -X POST http://127.0.0.1:8787/api/generate
```

## Keys

| Key | Purpose |
|-----|---------|
| `GEMINI_API_KEY` | Lyria 3 |
| `DEEPSEEK_API_KEY` | DJ brain |

## Siblings

- `tinqs-ltd/docs` — hub
- `tinqs-ltd/audio` / `music` — Ariki production audio (FMOD, stems)
- `tinqs/studio` — inference proxy for DeepSeek