Weaviate

Weaviate is an open-source vector database with the broadest "batteries-included" story of any vector engine. Its module system can pull embeddings from OpenAI, Cohere, Hugging Face, or local models, and its generative-* modules let you run RAG generation inside the database, returning grounded answers in a single query.

For agents, the killer feature is multi-tenancy: Weaviate's tenant-per-collection model can scale to hundreds of thousands of tenants on a single cluster, with per-tenant activation and offloading to cold storage. Hybrid search (BM25 + vector with configurable alpha) is first-class, not an add-on.

Weaviate publishes an official MCP server that exposes collection management and search to MCP-aware agents. The Serverless Cloud tier separates storage from compute for cost-efficient bursty workloads, and the BYOC (bring-your-own-cloud) option satisfies stricter data residency requirements without giving up the managed experience.

The tradeoff is resource appetite — Weaviate is heavier than Qdrant or Chroma in memory and CPU, and the module system rewards investment in configuration to get right.

Agent-specific features:

• Tenant-per-collection isolation that scales to 100k+ tenants per cluster
• Hybrid search (BM25F + vector) with configurable fusion
• Vectorizer and generative modules: embeddings and RAG happen inside the database
• Official MCP server for agent-driven collection and search operations
• Serverless Cloud with separated storage and compute, plus BYOC for data residency