Vearch

Vearch is an open-source, cloud-native distributed vector database for efficient similarity search of embedding vectors in AI applications. It combines vector search with scalar filtering and offers RESTful APIs plus Python, Go, Java and Rust SDKs. Source code is released under the Apache License 2.0.

What It Is

Vearch stores and indexes vectors and scalar fields, then retrieves them through hybrid search that pairs vector similarity with filter conditions. The project states it can search millions of objects in milliseconds and supports replication and elastic scale-out.

Architecture

A Vearch cluster has three components. The Master handles schema management, cluster-level metadata and resource coordination. The Router exposes the RESTful operations upsert, delete, search and query, and handles request routing and result merging. PartitionServers host document partitions with raft-based replication, and run Gamma, the core vector search engine built on faiss, which stores, indexes and retrieves vectors and scalars.

Use as an AI memory backend

Vearch can serve as a vector store or memory backend for AI frameworks. Integrations are provided for LangChain, LlamaIndex, Langchaingo and LangChain4j. The README also describes a visual search demo for indexing large image collections, which needs an image retrieval plugin for object detection and feature extraction.

Deployment options

Deployment options include Kubernetes via a Helm chart, Docker Compose in standalone or cluster mode, Docker, and compiling from source.