Weaviate

Objects and their vectors sit side by side in Weaviate, an open source cloud-native vector database. A single query interface combines vector similarity search with keyword filtering, retrieval-augmented generation and reranking, which spares a search pipeline from stitching those steps together across separate services.

Vectors reach the database in one of two ways. Integrated model providers such as OpenAI, Cohere and HuggingFace vectorize objects at import, or you import embeddings generated elsewhere, and the choice is made per collection when the schema is created.

The parts that matter in production belong to the database itself.

• Multi-tenancy: tenants are isolated inside a single deployment instead of spread across separate instances.
• Replication: data is replicated across nodes for availability.
• RBAC authorization: role-based access control governs who may read and write what.
• Client libraries: a Python client and clients for other languages speak to the same API.
• Local embedding models: a lightweight vectorizer runs beside the database in Docker, so no external provider is required to start.

Deployment runs from a Docker compose file on a laptop through Kubernetes, the managed Weaviate Cloud, and the AWS and GCP marketplaces. Common uses named by the project include retrieval-augmented generation systems, semantic and image search, recommendation engines, chatbots and content classification.