RuVector
github.com/ruvnet/ruvector- Category
- AI Tools
- Rank
- No. 915Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- ruvnet
- GitHub
- 4.5k stars
- Latest release
- ruvector-v0.2.40
- Date
About
rUv's high-performance, real-time, self-learning vector graph memory database written in Rust — designed for agentic AI memory at scale.
What it does
RuVector gives agents durable recall across processes and sessions. It embeds text locally or through external providers, stores vectors with metadata and relationships, then retrieves memories by similarity, filters, time, or graph traversal. Feedback and recorded outcomes can update ranking or learning state.
Why it's ranked here
The strongest case is its unusually complete local memory loop: encoding, durable retrieval, graph context, feedback, compaction, snapshots, and several integration surfaces. The verdict is mixed because that breadth also creates a sprawling system, while the unified four-type memory manager and cross-type consolidation remain incomplete.
What's good
The default path needs neither a database server nor an API key. Stored vectors, metadata, configuration, and searchability survive process restarts. Users can choose exact or approximate search, combine semantic and exact-term retrieval, apply metadata filters, and record outcomes without having ordinary reads silently change learned weights.
Tradeoffs
Each store must keep one embedding model and vector dimension, so changing an established store requires re-embedding. The unified working, episodic, semantic, and procedural layer is memory-resident, with unfinished cross-type consolidation. The repository covers many specialized subsystems, increasing selection and build complexity. Shared hosted memory also creates a separate data boundary.
How to use it well
Use it for Rust or Node agents that need project-local semantic recall, structured metadata, relationships, and explicit feedback across sessions. Start with the local command line or embedded store, then add graph, temporal, or shared-memory features only when required. It does not decide what deserves storage, which evidence is trustworthy, when memories expire, or what recalled context may control.
Technical notes+
README.md documents local ONNX embeddings, persistent VectorDB storage, lower-is-closer distance scores, HNSW and flat indexes, graph retrieval, snapshots, and explicit feedback learning. Cargo.toml defines a very large Rust workspace spanning CLI, server, Node, WASM, graph, PostgreSQL, replication, MCP, and specialized retrieval crates. package.json exposes Cargo-backed build, test, benchmark, CLI, and MCP scripts, with Node 18 or newer specified. docs/sdk/INDEX.md says the Python SDK is a planning artifact without a pyproject.toml, PyO3 dependency, or implementation. crates/mcp-gate/src/server.rs and crates/mcp-brain/src/server.rs implement JSON-RPC 2.0 MCP servers over stdio. crates/ruvector-graph-transformer-node/index.js selects native bindings across Android, Windows, macOS, FreeBSD, and several Linux architectures.
Observed
- License
- MIT
- Primary language
- Rust
- Packaging
- Rust crates and an npm package; Node requires version 18 or newer
- Interfaces
- Rust library, Node.js library, CLI, server, WASM, PostgreSQL extension, and stdio MCP servers
- Platform bindings
- Native Node binding loader covers Android, Windows, macOS, FreeBSD, and Linux across multiple architectures
- Python surface
- The Python SDK is planning documentation only; no first-party Python package implementation is present
Read from README.md, Cargo.toml, package.json, docs/INDEX.md, docs/sdk/INDEX.md, crates/ruvector-graph-transformer-node/index.js, crates/mcp-gate/src/lib.rs, crates/mcp-brain/src/lib.rs, crates/mcp-gate/src/main.rs, crates/mcp-brain/src/main.rs, crates/hailort-sys/src/lib.rs, crates/mcp-gate/src/server.rs, crates/mcp-brain/src/server.rs, crates/ruos-thermal/src/lib.rs, crates/ruvector-cnn/src/lib.rs.
What it can do
Store vector embeddings in graph database
Vector embeddings and relationships → Structured graph memory storage
Perform real-time vector similarity search
Query vector → Similar vectors with relevance scores
Learn and adapt memory patterns automatically
Usage patterns and data interactions → Optimized memory organization
Scale vector operations across distributed systems
Large-scale vector datasets → Distributed processing results
Manage AI agent memory states
Agent context and conversation history → Persistent memory context
Execute graph traversal queries on vector data
Graph query parameters → Connected vector nodes and paths
Process high-throughput vector operations
Concurrent vector requests → Real-time processing results
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.