AI Discovery

AI Discovery

Discover (menu key 3) lets you describe what you want to hear in plain words and get back a list of real songs to play. Type "calm piano for focus" or "90s hip hop classics", pick a song from the list, and it streams from YouTube like any other track.

The feature is off until you add an API key. tunetape ships no key and makes no AI calls on its own. You bring your own, from Anthropic, OpenAI, any compatible gateway, or a local model server.

Setup in two minutes

1. Get a key. For the default provider that is an Anthropic API key.
2. Open Settings → 3. AI discovery and press 2: a small box opens to paste the key into (it is hidden as you paste). It is stored in your local config.json and only its last four characters are ever shown.
3. Press t to Test connection. The loading box ends with a green ✓ Connected and the model that answered, and a notification repeats the result when you're back on the settings.
4. Back on the welcome screen, press 3 and type a request.

If you would rather not save the key, leave the field blank and export ANTHROPIC_API_KEY (or OPENAI_API_KEY for the openai provider) in your shell instead. A saved key wins over the environment variable.

The settings, one by one

Setting	Key	What it does
Provider	1	Flips between anthropic (Anthropic Messages API) and openai (any /v1/chat/completions endpoint: OpenAI, Ollama, LM Studio, vLLM, gateways)
API key	2	The key sent with each request. Submitting the box empty keeps the current key; with no key saved, the environment variable for the provider is used
Model	3	Any model id the provider accepts. The box opens with the current one; clear it to reset to the default
Endpoint	4	Base URL. tunetape POSTs to <base>/v1/messages (anthropic) or <base>/v1/chat/completions (openai). Clear it to reset to the default
Auth	5	Anthropic only: send the key as x-api-key (the official API) or as a Bearer token (many proxies and OAuth gateways). The openai provider always uses Bearer
Clear saved key	c	Forget the stored key
Test connection	t	One tiny request to confirm key, model and endpoint work together

Press a row's key, or move with the arrows and press Enter. In the text boxes, Enter saves and esc cancels.

Defaults

Provider	Model	Endpoint
anthropic	claude-haiku-4-5-20251001	https://api.anthropic.com
openai	gpt-4o-mini	https://api.openai.com

The defaults are small, fast models: a Discover request is a few hundred tokens, so it costs a fraction of a cent. Any model id works if you want a bigger one.

Local models (Ollama, LM Studio and friends)

Anything that speaks the OpenAI chat-completions protocol works, so you can run Discover fully offline:

1. Provider → openai
2. Endpoint → your server's base URL, for example http://localhost:11434 for Ollama or http://localhost:1234 for LM Studio
3. Model → a model you have pulled, for example llama3.2
4. API key → any non-empty text. Local servers ignore it, but tunetape needs something there to consider the feature configured
5. t to test

Small local models sometimes return songs that do not exist. tunetape only lists what the model says; if a pick does not play, try the next one.

Gateways and proxies

Point Endpoint at the gateway's base URL and keep the provider matching the protocol it speaks. For an Anthropic-compatible gateway that expects Authorization: Bearer ..., switch Auth to Bearer. The anthropic-version header is always sent with the anthropic provider; compatible gateways ignore it.

How a request works

• Your request text goes to the model with a short system prompt asking for real, existing songs (artist and title), up to 12 of them.
• The reply is turned into a list. Each entry becomes a YouTube search (<artist> <title>) that yt-dlp resolves when you pick it, so yt-dlp is required for Discover.
• The request itself is saved to Recently played (tagged AI · search) so you can re-run it later for a fresh list, and each song you play from it is saved as its own AI entry.
• Before you type, Discover runs a quick connection check (skipped once it has passed this session). If it fails you see why, with s as a shortcut into the AI settings.

Privacy

Only the text you type into Discover (plus a short fixed system prompt) is sent to the provider you configured. Your listening history, links and settings never leave your machine. The API key is stored in plain text in config.json in your data folder (see Settings); use the environment variable instead if you prefer not to keep it on disk.

Common errors

Message	Fix
AI discovery isn't set up yet	Add a key (Settings → AI discovery, or the environment variable)
Auth / 401 / 403	Wrong or expired key, or the wrong Auth mode for your endpoint
Model not found / 404	The model id does not exist at that endpoint. Blank it to reset to the default
Connection error / timeout	The endpoint is unreachable or slow (tunetape waits up to 45 seconds). Check the URL and that a local server is running

More in Troubleshooting.