Consider extending compass to use vector database techniques…

TurboVec is a memory-efficient vector search index for RAG and similarity search, prioritising privacy and local deployment. It implements Google Research's TurboQuant algorithm for aggressive vector compression (16x reduction on typical embeddings), achieves 12-20% faster search than FAISS on ARM, and supports online ingestion without retraining. Written in Rust with Python bindings and hand-optimised SIMD kernels (NEON/AVX-512BW). Integrates with LangChain, LlamaIndex, Haystack, and Agno. MIT licensed; available via PyPI and crates.io.

Using turbovec could improve compass search relevance beyond the current FTS approach — semantic similarity rather than keyword matching — while remaining fully local with no managed services.

This page is a capture in the inbox bucket of the product backlog — a pre-sprint idea, not yet pulled into a sprint as a story.