Skip to content
mini-tui

Providers and models

Bring your own model keys to mini-tui: supported providers, how a key is tested and stored, how a model is chosen, and which key each run receives.

1 min read · Edit this page

mini-tui talks to models directly, with no proxy layer in between. You bring the key.

Supported providers

Xiaomi MiMo, DeepSeek, OpenCode Go, Z.AI, MiniMax (global and China), OpenAI, Anthropic, Moonshot, Zhipu, Groq and OpenRouter. Any OpenAI-compatible endpoint also works through the generic openai/ slot.

How a key is handled

  1. You paste the key in /connect or Settings, Providers.
  2. mini-tui asks the provider for its model list, then sends one real request through the agent's own model layer. A passing test means a run will work with this key.
  3. Only then is the key saved, to ~/.config/mini-tui/providers.json with 0600 permissions. A typo never replaces a working connection.
  4. The web API never returns a key. Reads carry a masked hint such as sk-…a1b2.

Which key a run gets

A run receives the credentials of its own model's provider only. Choosing deepseek/deepseek-chat puts the DeepSeek key in that run's environment and nothing else.

Choosing a model

Use /model, or -m <id> on the command line. The last model you pick anywhere becomes the default for new chats. In the web app the picker is searchable and grouped by provider, and a model that is not listed can be typed.