Zvec: In-Process Vector Database (Alibaba)

Purpose

Lightweight, open-source vector database engineered for embedding directly into applications. Developed and battle-tested within Alibaba Group.

Key Features

• Performance: Searches billions of vectors in milliseconds
• Simplicity: Install and begin searching within seconds; pure local operation without servers
• Vector Support: Handles both dense and sparse embeddings with native multi-vector query capabilities
• Hybrid Search: Combines semantic similarity with structured filtering
• Data Durability: Write-ahead logging (WAL) ensures persistence even during crashes
• Concurrent Access: Multiple processes can simultaneously read collections; writes maintain single-process exclusivity
• Portability: Functions as in-process library, compatible with notebooks, servers, CLI tools, and edge devices

Installation

Available for Python 3.10-3.14 (pip install zvec) and Node.js (npm install @zvec/zvec). Supports Linux (x86_64, ARM64), macOS (ARM64), and Windows (x86_64).

Technical Stack

Primary implementation in C++ (79.5%), with Python bindings (7.8%) and SWIG language integration (7.7%).