Milvus vs Vearch

Milvus is an open-source vector database from Zilliz for GenAI and scalable vector search. It supports approximate-nearest-neighbor, filtered, range, hybrid, and full-text search, along with reranking, fetch, and query operations. Metadata filtering, sparse vectors, and index types including HNSW, IVF, FLAT, DiskANN, and SCANN are documented. Deployment choices are Milvus Lite as a Python library, Standalone as a single-machine server, and Distributed on Kubernetes. Documentation recommends Lite for up to a few million vectors, Standalone up to 100 million, and Distributed for 100 million to tens of billions. Lite can run locally without a cloud service for private or sensitive data; an online upgrade from Standalone to a cluster is not currently supported. Official interfaces include REST plus Python, Go, Java, Node.js, and C++ APIs or SDKs. C# is contributed by Microsoft and Rust is under development. Security documentation covers authentication, TLS for gRPC and REST, and role-based access control at collection, database, and instance levels. The Apache 2.0 software is free, and Zilliz Cloud is offered as a managed option.

Vearch is a free, open-source, cloud-native distributed vector database for similarity search in AI applications. It searches embedding vectors and can combine vector queries with scalar filtering for metadata conditions. The project describes replication and elastic scale-out and says it can retrieve vectors among millions of objects in milliseconds. Its Router exposes REST operations for upserting, deleting, searching, and querying data. Client SDKs are listed for Python, Go, Java, and Rust, with integrations for LangChain, LlamaIndex, LangChainGo, and LangChain4j. Deployment options include Kubernetes through Helm, Docker Compose in standalone or cluster mode, Docker, and compilation from source. PartitionServer uses Raft-based replication for document partitions, while Gamma is a Faiss-based vector search engine. Vearch is licensed under Apache License 2.0 and supports Linux and macOS self-hosted deployments. Community support is available through GitHub Issues, an email discussion group, and Slack. The VisualSearch demo requires an image retrieval plugin for object detection and feature extraction.