RuVector

Why we included this project

Agents that need to remember things between sessions can use RuVector as a real storage layer rather than a demo. It runs entirely on your machine, with no database server and no API key. The local embedding model is cached on first use, so retrieval costs nothing per query and sensitive data never has to leave your control. On top of vector search it layers graph relationships and explicit feedback learning, so recalled context improves from recorded outcomes instead of just sitting there. With Node.js and Rust bindings plus CLI hooks it slots into existing agent loops quickly, and while some typed memory managers are still maturing, the general-purpose vector store is the safest place to start for production work.