What's New
- OpenAI-compatible embedding provider — Use any endpoint that implements the
/v1/embeddingsAPI (llama.cpp, llamaswap, LiteLLM, vLLM, etc.). Configure a custom URL, model name, and optional API key in Settings → Embedding Provider. (#13) - Improved embedding cache invalidation — Changing the embedding URL or model now correctly invalidates cached vectors, including across app restarts.