hyperdb

A hybrid graph + vector database engine, in Rust.

Property graph and HNSW approximate-nearest-neighbor search in one embeddable engine — a hybrid query pipeline, SIMD distance kernels, and zero dependencies on external services.

---

What it is

hyperdb unifies two things most stacks keep apart — a property graph (nodes, edges, attributes) and vector search (HNSW ANN) — behind one query pipeline, so you can traverse relationships and search embeddings in the same engine. It is embeddable (a Rust library, no server to run) and depends on no external services.

It is the graph + knowledge layer behind CadFuture, the Charlot Lab's LUT-first CAD engine.

Features

• Property graph — typed nodes and edges with attributes.
• Vector ANN — HNSW and DiskANN-style indexes for high-dimensional similarity search.
• Hybrid query pipeline — combine graph traversal and vector search in one query, with a small query DSL.
• SIMD distance kernels — fast vector distance, with optional GPU paths.
• Quantization — compact vector storage.
• Change-data-capture and encryption at rest (AES-GCM, Argon2 key derivation).
• Embeddable & dependency-light — no external database or service required.

Use

[dependencies]
hyperdb = { git = "https://github.com/dcharlot-physicalai-bmi/hyperdb" }

cargo build --release
cargo test           # includes property tests (proptest)

Status

An active research-infrastructure project of the Charlot Lab at the Institute for Physical AI, Bailey Military Institute. Under active development; interfaces will change.

License

MIT © David Jean Charlot — see LICENSE.

---

_{The Charlot Lab · Institute for Physical AI · Bailey Military Institute}