#1818th of 23 in Vector databases
usearch
Proximity-graph nearest-neighbor index with bindings for 14 languages
- Stars
- 4.4k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Last release
- Oct 2026
- Language
- C++
Missed the latest GitHub check; these facts may be older.
Overview
USearch is a header-only C++20 library for approximate nearest-neighbor search over dense vectors, with experimental engines for sequences, 3D shapes and geospatial sets. Indexes can live in RAM, GPU VRAM or NVMe and share one checkpoint format. It also offers MaxSim multi-vector scoring and graph-selected sparse products. ClickHouse, DuckDB, ScyllaDB and YugabyteDB embed it.
Who it is for: Engineers embedding fast vector search into databases or applications
Strengths
- Header-only C++20 core with no external dependencies; one C ABI under all bindings
- RAM, VRAM and NVMe residencies exchange portable SafeGraphs checkpoint files
- Embedded in ClickHouse, DuckDB, ScyllaDB and YugabyteDB for vector search
- Compiles up to 11 storage types (bf16, i8, u1, e4m3 and more) and uses 128-bit keys
Weaknesses
- Library only: no server, REST API or built-in clustering
- Sequence, shape and geospatial engines are experimental and exposed only via C/C++
- Filtered search is missing in the JavaScript and Java bindings
- Geometry and storage types are fixed when the index opens
What it needs
- GPU optional
Also in Vector databases
See all 23| Rank | Project | Score |
|---|---|---|
| 1 | MilvusDistributed vector database with dense, sparse and hybrid search at scale | 90 out of 100 |
| 2 | QdrantRust vector database with payload filtering, REST and gRPC | 85 out of 100 |
| 3 | MeilisearchRust search engine with typo tolerance, filtering and hybrid semantic search | 84 out of 100 |
| #4 | faissC++ library for similarity search and clustering of dense vectors | 81 out of 100 |
| #5 | leannVector index that recomputes embeddings on demand to cut storage | 80 out of 100 |