RuVector

Use it when

• A coding or task agent forgets decisions and outcomes between sessions and you want file-backed semantic memory it can write to and recall from later processes
• You need embedded vector storage with local ONNX embeddings inside a Node.js or Rust service, without standing up Qdrant, Milvus, or a cloud vector API

What it solves

Not the fit when

• Automatic memory curation; the application decides what is trusted, worth remembering, when it expires, and which actions recalled context may influence
• Using collections or metadata filters as an authorization boundary
• Treating embeddings or recalled content as harmless; embeddings remain sensitive derivatives and recalled context is untrusted input until policy and action checks pass
• Fully offline first use without preparation; the default semantic path downloads all-MiniLM-L6-v2 on the first semantic command
• Large stores requiring fast cold start; persisted HNSW databases currently rebuild the index on open
• Production network replication; Raft transport and snapshot installation are incomplete
• Assuming npm installation activates every capability shown by the monorepo
• Relying on core quantization for physical compression; the option is persisted but not applied to core storage or indexes
• hosted multi-node vector search cluster
• turnkey RAG pipeline over documents
• automatic memory curation without application logic

Install

For the CLI memory path, run npx ruvector hooks remember/recall directly; for an embedded Node.js service, install with npm install ruvector. No database server or API key is required.

Invoke

CLI: npx ruvector hooks remember --semantic --type decision "fact" then npx ruvector hooks recall --semantic --top-k 3 "query"; Node: new VectorDB({dimensions: 384, storagePath: './agent-memory.db'}) with OnnxEmbedder for local embeddings

Alternatives

No reviewed alternatives recorded yet.