Blog
hvec is a Rust command-line tool for testing claims about compressing embedding vectors: what a codec saves in bytes, what it costs in retrieval quality against relevance judgments, and whether that loss changes the answers a chat model gives. It ingests documents, stores embeddings through a codec, retrieves in the compressed domain, asks a model, and records every stage. The benchmark is a matrix of embedding model × codec × payload stage × chat model × dataset.
- Code and README: github.com/SwineCoder101/hvec
- Experiment write-ups: experiments
- How to contribute: CONTRIBUTING.md
Posts below cover progress, design notes and results.
Posts
-
Similarity on encrypted vectors: the cost model, before any code
The roadmap’s last track is similarity search on encrypted embeddings: a codec whose compressed-domain operation is exact, where the thing it buys is not bytes but privacy. This note works out what such a codec costs before writing any of it, because the arithmetic decides whether it belongs in hvec’s
Codectrait at all. The short answer is that it does not, yet, and the reasons are instructive. -
Two layers of compression, and which one the agent never sees
An embedding is already a lossy codec. hvec adds a second one. Here is what each layer compresses, how they interact, and why the agent reads text, not codes. -
hvec: measuring vector compression where it matters
Why hvec exists, what milestone 1 ships, and what comes next.
subscribe via RSS