Zilliz Cloud

Zilliz Cloud is a managed vector database built on open-source Milvus, for developers adding RAG, semantic search or recommendations to their apps.

Zilliz Cloud stores embeddings and answers similarity queries over them, so an application can fetch the passages, products or images closest in meaning to a request. Its engine is Milvus, the Apache 2.0 open-source vector database, run here as a fully managed service. Zilliz Inc., an independent company in Redwood City, California, builds and operates it. In June 2026 Zilliz put Vector Lakebase, a major Zilliz Cloud release, into public preview. AUTOINDEX selects the approximate nearest-neighbor index automatically, and memory-mapped and hybrid disk modes keep indexes on disk as well as in memory. Filtered search applies metadata conditions before the similarity step, while sparse-dense hybrid search blends BM25 keyword scoring with dense vectors through RRF or weighted rankers. Embedding functions turn text into vectors through hosted or third-party models, and rankers from Cohere, Voyage AI and Hugging Face reorder the top results. One collection can hold several vector fields for text, images and audio, and a Partition Key keeps each tenant's data apart. Snapshots and cluster backups restore collections, custom cluster roles grant privileges down to a single collection, and bulk import loads Parquet, JSON or NumPy files from cloud object storage. The service suits developers adding retrieval-augmented generation, semantic search or recommendations to a product who want Milvus without running it themselves. Teams prototype on Serverless clusters, then pick a Dedicated cluster type: performance-optimized for low latency, capacity-optimized for larger datasets, or tiered storage for the largest sets. Larger companies get SAML single sign-on, audit logs, private endpoints and customer-managed encryption keys. Regulated teams can use HIPAA-eligible deployments, and a bring-your-own-cloud option runs the service on infrastructure the customer chooses. One job is the retrieval step behind a chatbot: documents become embeddings, sit in a collection and come back by meaning when a user asks a question. Recommendation features use the same method to suggest products similar to the one on screen, and SaaS vendors serve every customer from one collection while keeping their vectors separate. Clusters run on AWS, Google Cloud or Azure. With On-demand Compute, query jobs and index builds run over external data without copying it into a cluster. It fits any team whose search must match what a request means rather than only the words it contains.