Weaviate

Weaviate is an open-source vector database for RAG, hybrid search and agent memory, run self-hosted or managed on Weaviate Cloud with usage-based plans.

Weaviate gives AI applications a place to store embeddings and search them by meaning, so a chatbot or a search box can find the right passage even when the words differ. It is developed by Weaviate B.V., an independent company in Amsterdam. The engine's source code is public on GitHub, mostly under the BSD-3-Clause licence. Teams run it on their own servers, as a Kubernetes package in their own VPC, or as the managed Weaviate Cloud service. The vector database stores objects in collections and indexes their vectors with HNSW, flat, dynamic or the disk-based HFresh index. Hybrid search blends BM25 keyword scoring with vector similarity in one query, and metadata filters apply before the vector search rather than after it. Rotational, product, binary and scalar quantization shrink the index held in memory. Weaviate Embeddings generates vectors from text and images with models hosted in Weaviate Cloud, while reranker integrations reorder the top results. Query Agent turns plain-language questions into database operations, and Engram builds a memory that adapts to each user over time. The natural buyers are engineering teams shipping retrieval-augmented generation, semantic search or AI agents. SaaS companies get multi-tenancy, where each tenant sits on its own shard and never sees another tenant's data. Larger systems split collections into shards across nodes and replicate them, so queries keep answering when one node goes down. Role-based access control scopes each user and API key to particular collections, tenants and operations. Companies with strict data rules can keep the whole database inside their own cloud account instead of using the managed service. Day to day, Weaviate grounds an assistant's answers in a company's own documents and powers catalog search that understands intent as well as keywords. Embeddings and rerankers can come from providers such as OpenAI, Cohere, Mistral or Hugging Face when a team prefers those models to the hosted ones. Managed clusters run on AWS, Google Cloud or Azure, and backups go to S3, Google Cloud Storage or Azure Storage for restore. A team that wants one open-source system for vectors, retrieval and user memory, with the freedom to move it between its own hardware and the managed cloud, is the team it is built for.