alibaba /zvec

A lightweight, lightning-fast, in-process vector database

★ 16,244⑂ 1,022👁 70⑃ 675● 200▤ 16
C++Apache-2.0Created 2025-12-05zvec.org
Download: v0.8.1v0.8.1 ↓0.8.12026-11-06
+3
Daily
16,244
stars
1,022
forks
70
watchers
675
prs
200
issues
16
releases

README

English | 中文



  
    
    
  



  
  
  
  



  



  🚀 Quickstart  |
  🏠 Home  |
  📚 Docs  |
  📊 Benchmarks  |
  🔎 DeepWiki  |
  🎮 Discord  |
  🐦 X (Twitter) 


Zvec is an open-source, in-process vector database — lightweight, lightning-fast, and designed to embed directly into applications. Battle-tested within Alibaba Group, it delivers production-grade, low-latency and scalable similarity search with minimal setup.

 [!Important]
 🚀 v0.8.0

 - Broader platform support: Added Intel Mac and Windows ARM64 builds, wheels, and SDKs; extended DiskANN to Windows, Android, and iOS, and RaBitQ to Windows x86_64.
 - Chinese FTS ready to use: Embedded the default Jieba dictionary and HMM model, so Chinese tokenization needs no separate dictionary files or path configuration.
 - Lower index build costs: Reduced peak memory during IVF / DiskANN builds and merges, and accelerated HNSW RaBitQ construction through vector reuse and batched lookups. Build and nested training tasks now enforce configured thread limits more strictly.
 - Faster distance computation: Extensively refactored quantizers and distance calculations to benefit multiple
    quantization types and metrics, with major improvements to INT4 kernels and Linux ARM64 performance.
 - Faster search: Added faster Uniform quantizers and optimized Vamana graph layout and prefetching. Uniform quantization avoids per-point score reconstruction, though a shared min/max range can increase quantization error and reduce recall.
 - Bug fixes and improvements: Added more flexible UTF-8 document IDs up to 1,024 bytes; fixed concurrent reads during segment switches, index rebuilding with deleted rows, and correctness issues in LIKE and floating-point range filters.

 👉 Read the Release Notes

[^query-performance]: Measured on GIST 1M with L2 distance, INT8 quantization, and HNSW on a 16-core Linux ARM64 system.

 💫 Features

- Blazing Fast: Searches billions of vectors in milliseconds.
- Simple, Just Works: Install and start searching in seconds. Pure local, no servers, no config, no fuss.
- Dense + Sparse Vectors: Support dense and sparse embeddings, multi-vector queries, and a rich selection of vector index types that scale from memory to disk.
- Full-Text Search (FTS): Native keyword-based full-text search — query string fields with natural-language or structured expressions.
- Hybrid Search: Fuse vector similarity, full-text search, and structured filters in a single query for precise results.
- Durable Storage: Write-ahead logging (WAL) guarantees persistence — data is never lost, even on process crash or power failure.
- Concurrent Access: Multiple processes can read the same collection simultaneously; writes are single-process exclusive.
- Runs Anywhere: As an in-process library, Zvec runs wherever your code runs — notebooks, servers, CLI tools, or even edge devices.

 📦 Installation

Zvec offers official SDKs across multiple languages:

| SDK | Install / Dependency | Source |
| --- | --- | --- |
| 

Top issues

Discussions

Latest release

v0.8.1 v0.8.1 2026-11-06
🚀 Zvec v0.8.1 Release Notes

 📦 New Features
+ PQ for HNSW and IVF: Added PQ4/PQ8 and optional OPQ, covering training, search, persistence, and subsequent updates. (833)
+ Broader platform support: Enabled RaBitQ on Linux, macOS, Android, and iOS ARM64, plus Intel Mac. DiskANN now supports Intel Mac and Windows ARM64. (832, 874)
+ Grouped search in the C API: Added grouped vector query and result-management APIs. (845)

 🚀 Performance Improvements
+ Lower build memory: Reduced vector copying and temporary storage in DiskANN, IVF, and selected HNSW construction paths. (853, 877, 879, 88

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON