# ANNex > Rust-native vector search engine with hybrid dense-sparse retrieval, multivector support, and a recall-latency frontier that outpaces hnswlib up to 96% recall. ANNex is an embeddable vector search engine written in Rust. It implements HNSW for approximate nearest neighbor search with payload filtering, BM25/sparse retrieval, hybrid dense-sparse fusion (RRF), multivector / ColBERT-style late interaction, snapshots and a write-ahead log. Embeddings are supplied by the caller. ## Quick facts - Version: 0.3.0 (latest release; see https://annexsearch.dev/changelog.md) - License: Apache-2.0 OR MIT - Language: Rust (Python bindings: ANNexDB on PyPI) - Install: `cargo add annex` (Rust) or `pip install ANNexDB` (Python) - Headline benchmark: 0.18 ms p50 latency at 0.836 recall, 5,424 QPS (m16+rcm+sq8, ef=32, single thread; NYT256, 290K docs, 256-D angular, EC2, 2026-10-01) - Hybrid retrieval quality: macro nDCG@10 0.440 (ANNex hybrid, 5 BEIR corpora), parity with Qdrant and LanceDB, not a proven advantage - Pareto vs hnswlib: ANNex configs are faster at matched recall through 0.961; at 0.971 recall hnswlib is faster ## Capabilities - Dense vector search (cosine, dot product, Euclidean) - Sparse / BM25 retrieval - Hybrid dense-sparse fusion (RRF) - Multivector / late-interaction (ColBERT) - Payload filters on arbitrary key-value metadata - Embedded Rust library (no server required) - HTTP server (annex-multivector) - Python bindings (ANNexDB on PyPI) - Snapshot persistence and WAL replay - SQ8 and PQ quantization ## What it does NOT do (v0.3.0) - No managed cloud or hosted service - No distributed or sharded mode - No GPU acceleration ## When to choose it - You want an embeddable library: no server process, no network hop, integrated directly into your Rust binary. - You need dense and sparse retrieval in one index, with hybrid fusion (RRF), payload filters, or multivector / ColBERT-style late interaction. - You want snapshots and a write-ahead log built in, with Python bindings (NumPy batch search) for the read path. ## Pages (Markdown) - [Overview and quickstart](https://annexsearch.dev/index.md): what ANNex is, capability matrix, fit guidance, code examples - [Benchmarks](https://annexsearch.dev/benchmarks.md): every table with hardware, dataset, commit and date - [Changelog](https://annexsearch.dev/changelog.md): every release with notes, links to GitHub releases and package registries - [ANNex vs Qdrant](https://annexsearch.dev/compare/qdrant.md) - [ANNex vs FAISS](https://annexsearch.dev/compare/faiss.md) - [Everything in one file](https://annexsearch.dev/llms-full.txt) ## Data - [benchmarks.json](https://annexsearch.dev/data/benchmarks.json): All benchmark tables with full provenance (JSON) - [nyt256-pareto.csv](https://annexsearch.dev/data/nyt256-pareto.csv): NYT256 recall–latency Pareto (CSV) - [nyt256-head-to-head.csv](https://annexsearch.dev/data/nyt256-head-to-head.csv): NYT256 head-to-head vs FAISS, hnswlib, USearch (CSV) - [nyt256-ec2-topk10.csv](https://annexsearch.dev/data/nyt256-ec2-topk10.csv): NYT256 throughput on EC2, top-k=10 (CSV) - [beir-ndcg10.csv](https://annexsearch.dev/data/beir-ndcg10.csv): BEIR nDCG@10 vs LanceDB and Qdrant (CSV) ## Links - Source: https://github.com/rosharma719/ANNex - Docs: https://docs.rs/annex - Crate: https://crates.io/crates/annex - PyPI: https://pypi.org/project/ANNexDB/ - Benchmark policy: https://github.com/rosharma719/ANNex/blob/master/BENCHMARK_POLICY.md - Benchmark corrections: https://github.com/rosharma719/ANNex/blob/master/crates/annex-multivector/benchmark/RESULTS.md - Releases: https://github.com/rosharma719/ANNex/releases