ANNex
ANNex is a Rust workspace for vector and late-interaction retrieval.
| Component | Purpose | Documentation |
|---|---|---|
annex |
HNSW, payload filters, sparse retrieval primitives, snapshots and WAL | Rust API, operations |
annex-multivector |
Persistent hybrid retrieval, named vector fields, filtered context selection and an HTTP API | API and quickstart, durability |
annex-server |
Dense-vector HTTP benchmark adapter | source |
annex-py |
Native Python/NumPy snapshot search and threaded batches | Python installation and API |
Embeddings are supplied by the caller. Deployment limits and remaining launch work are tracked in the implementation plan.
Build and use
Use the toolchain in rust-toolchain.toml:
To embed the core library from a local checkout:
[]
= "0.2.0"
The tested Rust quickstart lives in the crate documentation. For the multivector server, follow its quickstart.
Development and evaluation
- Contribution and test workflow
- Benchmark commands and reporting policy
- Dataset setup, test configuration
- Snapshot construction, recall analysis
- Historical benchmark corrections
Runtime tuning uses the existing VECTORDB_* prefix; see .env.example.
Core diagnostics use the log crate. Applications supply their logging backend.
Benchmark artifacts and local logs are generated outside versioned source.
Contributing and license
See CONTRIBUTING.md, Code of Conduct, and security reporting.
Licensed under Apache-2.0 or MIT, at your option. Contributions are offered under the same dual license.