Skip to main content

​
Purpose

Pro needs a working provider and task mapping in Settings → AI. Chat and Embedding are separate mappings; configuring one does not configure the other.
Settings → AI: provider list, task mapping, test connection

Screenshot TODO · images/screenshots/ai-features/configuration.webp

​
How to set it up

  1. Add a provider from the list (OpenAI-compatible, DeepSeek, OpenRouter, Gemini, Ollama, LM Studio, and others). Fill the API key. For a self-hosted endpoint, fill the Base URL.
  2. Refresh discovered models. Tweak temperature inline if needed.
  3. Map default tasks (names follow the UI): summary, tags, chat; Embedding (semantic / RAG); translation.
  4. Test the connection, then run a real ⇧⌘A call from detail.
Keys go in the macOS Keychain. More: BYOK.

​
Embedding

With no usable Embedding: Chat configured and Embedding empty is the usual case of Pro without semantic search or RAG.

​
RAG prompts

In RAG workspace settings you can edit Q&A, planning, compression, and title templates:
  • Compare and copy the read-only defaults
  • Custom templates list supported placeholders; unlisted ones are not injected silently
  • Starcat does not rewrite saved custom prompts
If summary / tags / translation expose custom prompts, try them on one repo before Auto Tidy. Output language follows the UI by default, or you can set it separately. See AI settings.

​
Notes

  • The App Store build cannot launch your local CLIs. Local inference (Ollama and similar) must already be running.
  • Changing the Embedding model can misalign old vectors; rebuild when the UI says so.
Last modified on August 25, 2026