Self-hosted alternatives to embeddings

E

embeddings

Vector embeddings with HNSW indexing, sql.js persistence, and hyperbolic support

embeddings is an agent skill from the ruflo project for working with vector embeddings, used for semantic search and pattern matching. It stores vectors in a persistent cache built on sql.js, a SQLite build that runs as WebAssembly so it works across platforms, and it searches them with HNSW indexing, a graph-based index for fast nearest-neighbour lookup. The skill also supports a Poincaré ball model for hierarchical data, and it offers normalization with L2, L1, min-max and z-score methods, and chunking with configurable size and overlap, which is useful for splitting long documents before they are embedded. The skill's features table gives the claimed speedups, including a search that is 150 to 12,500 times faster with HNSW and 75 times faster with the agentic-flow ONNX integration. Those are the skill's own figures and depend on the data and hardware, so measure them on your own data. It has a list of commands for embedding and searching, and the skill says to use it for semantic search, pattern matching and similar lookups. It suits developers adding search over notes, code or past agent results to an application. The skill passes Gen Agent Trust Hub, Socket and Snyk on skills.sh, and it has about 1,200 installs. Install it with npx skills add https://github.com/ruvnet/ruflo --skill embeddings.