What’s new
- Vector store + agentic RAG — every note is embedded into a vector store, and Bird retrieves the most relevant notes by semantic similarity before answering. The retrieval step is what lets Bird answer with grounded citations rather than guesses.
- Semantic global search — type a phrase and Bower finds notes that mean the same thing, even if the wording is different. “The experiment with the funny gel” finds the right note without needing the title.
- Hybrid ranking — semantic and keyword scores are blended so exact matches still win when they should (“note titled 2026-02-04”) while meaning-based queries surface notes you’d never find by string match.