An open-source, in-process vector database, Zvec is designed for embedding lightweight, high-performance similarity search directly into applications. Built on Alibaba's battle-tested Proxima engine, it offers production-grade, low-latency search capabilities with minimal operational overhead, ideal for applications requiring rapid integration of vector search functionalities.

How It Works

Operating as an integrated library, Zvec avoids separate server processes, allowing seamless embedding within host applications. It utilizes Alibaba's Proxima engine for efficient similarity searches across billions of vectors in milliseconds. The system supports both dense and sparse vector types, enabling hybrid search queries that combine semantic relevance with structured filtering. Data persistence is guaranteed via write-ahead logging (WAL), ensuring durability even on application crashes. While multiple processes can read from a collection concurrently, write operations are single-process exclusive.

Quick Start & Requirements

Installation is straightforward via package managers: pip install zvec for Python (versions 3.10 - 3.14) or npm install @zvec/zvec for Node.js. Supported platforms include Linux (x86_64, ARM64), macOS (ARM64), and Windows (x86_64). Building from source is an option, with detailed guides available. Official quick-start guides, documentation, and benchmarks are accessible via provided links.

Highlighted Details

• Performance: Capable of searching billions of vectors in milliseconds.
• Simplicity: Designed for "Just Works" deployment, requiring no external servers or complex configuration.
• Vector Flexibility: Natively supports dense and sparse vectors, including multi-vector queries.
• Hybrid Search: Enables combining semantic similarity search with structured data filtering.
• Durability: Write-ahead logging (WAL) ensures data is never lost, even on process termination.

Maintenance & Community

Zvec is actively developed, with recent releases (v0.3.x as of April 2026) introducing initial Windows and Android support, performance optimizations like RabitQ quantization and CPU Auto-Dispatch, and expanded ecosystem integrations (C-API, AI Agent support). Community engagement is fostered via a Discord server and X (Twitter) presence. A contributing guide is available for developers interested in participating.

Licensing & Compatibility

The specific open-source license for Zvec is not explicitly stated in the provided README. As an in-process library, it is designed for direct integration into various applications, suggesting broad compatibility.

Limitations & Caveats

Write operations to a Zvec collection are exclusive to a single process, although concurrent read access from multiple processes is supported. Initial support for Windows and Android platforms, introduced in version 0.3.0, may still be maturing.