AI configuration

Add Providers, map task models, Embedding, and testing

Entry point

Settings → AI

Setup steps

1. Add a Provider

   Pick a type (OpenAI / compatible API, DeepSeek, OpenRouter, Gemini, Ollama, LM Studio, etc.; list is authoritative). Enter API Key and Base URL (for self-hosted or compatible endpoints).

2. Discover models

   Refresh Discovered models; adjust temperature and other row-level params if needed.

3. Default task mapping

   Assign Provider + model per task (labels follow UI):

   • Summary, tags, chat
   • Embedding (semantic search / RAG vectors)
   • Translation
4. Test

   Use Test connection; then run a real call on detail with ⇧⌘A.

Keys are stored in macOS Keychain. More: BYOK.

Embedding matters

Without a working Embedding model:

• Semantic search is unavailable
• RAG index cannot build the vector side

Chat configured but Embedding empty is the most common “I have Pro but semantic/RAG still fails” case.

Prompts (when exposed)

Some tasks allow custom system prompts. Changes affect tone and length; test on one repo before using with Auto Tidy.

Output language

Default follows UI language; you can set AI output language separately. See AI settings.

Related

• Auto Tidy
• FAQ · AI