vearch

Why we included this project

Teams building retrieval-augmented generation or semantic search features typically need a storage layer that can find similar embeddings fast, and Vearch is a solid fit: it's a distributed, cloud-native vector database rather than a single-node store. The cluster separates master, router, and partition server responsibilities onto different tiers, so metadata, request routing, and index data don't compete for the same resources, which keeps horizontal scaling and replication straightforward. Search supports IVFPQ, HNSW, and binary index types, and a single query can combine vector distance with filters on numeric or string fields. Official Python, Go, Java, and Rust clients, plus ready-made integrations for LangChain, LlamaIndex, and LangChain4j, let it drop into a common RAG stack without much glue code. For sub-second similarity search over millions of vectors, it's a practical choice as the persistence and retrieval layer.