Weaviate Cloud

Weaviate Cloud is the fully managed version of Weaviate, the open-source AI database. The underlying engine combines vector similarity search, BM25 keyword search, and structured property filters into a single hybrid query — which has quietly become the default retrieval pattern for production RAG and agentic search because pure-vector retrieval misses too many exact-match and faceted queries. On top of that core, Weaviate ships generative modules that let you call OpenAI, Cohere, Anthropic, Hugging Face, Voyage, Jina, or Google directly from the database, so a single GraphQL or REST query returns embeddings, top-k results, reranking, and a synthesized answer with citations. In 2026 Weaviate added Engram (now GA), a personalized AI memory layer, and a Query Agent that takes natural-language questions and decomposes them into typed Weaviate queries — turning the database into a more complete platform than just "vector store."

Pricing is one of the most explicit in the category. Engram is free forever for one cluster, 100K objects, 1 GB memory, 10 GB disk, with usage-metered Embeddings (2K req/day) and Query Agent (1K req/month) included. Flex starts at $45/month on pay-as-you-go for shared clusters with full DB features, RBAC, and 99.5% uptime — appropriate for prototypes and small production workloads. Plus jumps to $280/month on a prepaid annual contract and adds SSO/SAML, 30-day backup retention, and 99.9% uptime. Premium starts at $400/month and adds the choice of dedicated deployment, 99.95% uptime, global multi-cloud coverage across AWS/GCP/Azure, and enterprise support with as low as 1-hour Sev 1 response and a dedicated technical account team. For regulated workloads, Bring-Your-Own-Cloud is available so customers can run Weaviate inside their own VPC under the managed control plane.

Multi-tenancy is first-class and unusually well thought out: tenants can be active, inactive (frozen — kept on disk but unloaded from memory), or fully offloaded to cold storage. For SaaS apps with one collection per customer, that distinction is what keeps a per-tenant vector model economical at thousands of tenants. Compared to Pinecone (simpler API but less hybrid-search depth), Qdrant (excellent open source, smaller managed offering), Milvus (highest-scale workloads, more ops overhead), and Chroma (great DX, narrower production feature set), Weaviate's differentiation is the depth of hybrid search and the breadth of in-database generative modules.

The risks are typical for a managed vector database: cost can climb fast as object count grows; tuning hybrid-search weights and HNSW parameters still requires real expertise; and pay-as-you-go Flex pricing can surprise you under bursty embedding workloads. Before committing to a paid tier, run a representative workload against Engram, measure objects-per-dollar, and decide whether you really need dedicated infrastructure or shared Plus is enough.