zvec

zvec is an open-source, in-process vector database that scales to billions of vectors with sub-millisecond latency.

Zvec is an open-source, in-process vector database developed by Alibaba. It provides lightweight, fast vector search capabilities with sub-millisecond latency and supports billions of vectors. Users can install a single package for pure local usage without servers or complex configuration. It supports both dense and sparse embeddings, enabling multi-vector queries in a single call and hybrid search with structured filters. Data persistence is guaranteed through write-ahead logging, and the system allows concurrent reads with single-process exclusive writes. It is designed for developers running code in notebooks, servers, CLI tools, or edge devices.

zvec at a glance

Best for: Software engineer
Runs on: self-hosted, open source
Pricing: Free
Released: December 2025
Category: AI coding tool
Website: zvec.org

Key features

Sub-millisecond search latency scaling to billions of vectors
Pure local in-process library requiring no servers
Dense and sparse embedding support with multi-vector queries
Hybrid search combining semantic similarity and structured filters
Durable storage with write-ahead logging for persistence
Concurrent process access for simultaneous reads

Use it for

Scale vector search to billions of vectors locally
Perform hybrid search with semantic similarity and structured filters
Store embeddings durably with write-ahead logging
Run vector queries in notebooks, servers, CLI tools, or edge devices

Alternatives to zvec

TurboVec — Fits a 10 million document corpus in 4 GB of RAM and searches faster than FAISS.
ColdIVF — It reads under 1 percent of the index during cold queries.
seekdb — Unifies relational, vector, and text data for production-grade AI agents.
Milvus — Distributed vector search that scales to billions of vectors
Pg Vectorsearch — Provides faster querying and building than standard indexes.
HelixDB — Combines graph, vector, and full-text search in one engine.
Infinity — Achieves 0.1 milliseconds query latency on million-scale vector datasets.
Weaviate — Stores data objects and vectors for semantic and hybrid search capabilities.
Chroma — Object storage architecture with automatic query-aware data tiering and caching.
USearch — It uses SIMD instructions and optimized concurrency to outperform FAISS.
Glider — It combines crash-safe S3 durability with fast local SSD caching for vector search.
FastEmbed — It uses ONNX Runtime for high speed and does not require a GPU or PyTorch.

All zvec alternatives