Providers and models

Entering an API key

Settings → pick a provider → paste the key.

A set of mainstream providers is built in (DeepSeek / OpenAI / Claude / Gemini / Qwen / Kimi...), and you can also enter a custom address (any OpenAI-compatible gateway works).

Choosing a model

Settings → Model opens model management:

• Grouped by provider; each row shows the context length and price (free ones are green)
• The top right switches between "set as main model / set as small model"
• Selected ones show Main✓ / Small✓

You can also switch quickly from the Chat page: ☰ menu → model picker.

Main model and small model

• Main model: does the real work (writing code, analysis, decisions)
• Small model: does the chores (context compaction, summarization)

Selecting both saves the most money: chores go to the cheap small model, and only real work uses the expensive one. If you do not pick a small model, the main model does the chores itself.

Model fallback chain (desktop)

You can configure a list of models to try in order when the main one fails. Not available on the phone yet.

Other providers work too

As long as it is compatible with OpenAI's /v1/chat/completions, enter the address and key and you are set. Model management → Import can also bulk-import from the configs of tools like Claude Code / Codex / OpenCode.