Zvec
lastmod 2026-03-03created 2026-03-033 edits
Ultra-lightweight, ultra-fast in-process vector DB
About
- A vector DB that can be embedded directly in an application, designed to run large-scale similarity search with minimal setup
- Built on Alibaba's Proxima engine, providing production-grade scalability and low-latency search performance
- Supports both dense and sparse vectors, and can run multi-vector queries in a single call
- Hybrid search combines semantic similarity with structured filtering for precise search
- An in-process library that runs immediately with no server or setup, so it works anywhere: laptops, servers, CLI, edge environments, etc.
- Supports Linux (x86/ARM) and macOS, and installs easily in Python/Node.js environments
- Consists of a C++ core engine with SWIG and Python bindings, supporting high-performance computation and integration with various languages
- Optimizations such as unordered_map for O(1) average key lookup shorten index load time and improve memory efficiency
- Uses Int4/Int8 quantized operations to balance speed and accuracy when processing large numbers of vectors
- Apache-2.0 license
See also
- Vector Database
- Alibaba