Weaviate

Weaviate is an open-source vector database (Go) that stores objects alongside their vectors, so semantic search, keyword (BM25) search, and generative retrieval can live in one system. Its client modules can vectorize text automatically at import time using providers like OpenAI, Cohere, or Hugging Face, removing the need for a separate embedding pipeline.

It supports multi-tenancy, horizontal sharding, metadata filtering, and a GraphQL/REST API, with managed Weaviate Cloud as an alternative to self-hosting. The core database is BSD-3-Clause; enterprise-only code in the repo's wl/ directory is proprietary.

1. Install Docker and start Weaviate locally with the official compose setup: docker compose up -d
2. Install the v4 Python client: pip install -U "weaviate-client"
3. Connect: import weaviate then client = weaviate.connect_to_local() (pass headers={"X-OpenAI-Api-Key": key} if you use OpenAI vectorization)
4. Create a collection with automatic vectorization: client.collections.create(name="Article", vector_config=Configure.Vectors.text2vec_openai())
5. Insert objects with collection.data.insert({...}), then search with collection.query.near_text(query="...", limit=5) or run in-database RAG with collection.generate.near_text(...)