RyanCodrai/turbovec — reverse-engineered prompt

Build me a local vector search library for embeddings that is very memory efficient and fast. I want to be able to add lots of vectors, search for the closest matches, and save and reload the index from disk without sending any data to a hosted service.

Make the core in Rust for speed, but make it easy to use from Python with a simple API. The Python user should be able to create an index with a vector dimension and compression level, add vectors, search with top k, and persist it. Also add a version that lets people use their own stable numeric ids, delete by id, and search only inside an allowed list of ids for things like tenant filters or permissions.

Please include clear examples, tests, and basic benchmark scripts comparing memory use and search speed. Add docs for using it directly and for plugging it into common RAG workflows like LangChain, LlamaIndex, Haystack, and Agno. Look up current packaging docs online if needed.