v0.2.0

New

• Model detection per provider. baifo now maps each declared provider to its available models offline, and flags custom endpoints instead of guessing their model list.
• Smarter TUI autocomplete. Model suggestions are now scoped to the selected provider, and provider type suggestions are limited to what baifo actually supports.
• Tighter token control in filesystem tools. The search tool now has a configurable output cap, so broad queries no longer flood the context window.

Performance

• Faster model listing and TUI autocomplete thanks to a one-time catalogue index.

Breaking changes

• The openai-compatible and ollama provider types are gone. Declare any OpenAI compatible endpoint (Ollama, OpenRouter, vLLM) as type: openai with a url.
• Invalid auth modes now fail at startup instead of being silently ignored.

Documentation

• New configuration guide at docs/CONFIGURATION.md, plus an updated README.