alibaba/zvec
A lightweight, lightning-fast, in-process vector database
What it solves
Zvec provides a lightweight, high-performance vector database that can be embedded directly into an application as an in-process library. It eliminates the need to manage separate database servers or complex configurations while enabling low-latency similarity search across billions of vectors.
How it works
Zvec functions as an embedded library rather than a standalone server, running wherever the application code is executed (including notebooks and edge devices). It supports both dense and sparse embeddings and offers multiple index types that scale from memory to disk. To ensure data safety, it uses write-ahead logging (WAL) for persistence and allows multiple processes to read the same collection simultaneously.
Who it’s for
It is designed for developers who need production-grade similarity search and vector storage without the overhead of a separate server, particularly for applications running on servers, CLI tools, or edge devices.
Highlights
- Hybrid Search: Combines vector similarity, full-text search (FTS), and structured filters in a single query.
- In-Process Architecture: No server setup required; it embeds directly into the app for minimal latency.
- Broad Language Support: Official SDKs for Python, Node.js, Go, Rust, and Dart/Flutter.
- Advanced Indexing: Supports HNSW, HNSW-RaBitQ, and DiskANN, with features like random rotation quantization to boost recall.
- Full-Text Search: Includes a native FTS pipeline with a Unicode standard tokenizer and support for over 34 languages.