Qdrant

Qdrant is an open-source vector database written in Rust, known for raw performance and precise filtering. Its core abstraction is the collection of points — each point a vector plus a JSON payload — and its standout engineering is filterable HNSW: metadata filters are woven into the graph traversal itself, so heavily filtered queries stay fast instead of degrading the way post-filtering does. You can run it as a single Docker container, embed it in-process, cluster it for scale, or use the managed Qdrant Cloud, which includes a free tier.

In this module’s lineup, Qdrant is the performance-and-control option: fully open source under Apache 2.0 like Weaviate, but leaner — it doesn’t try to run your embedding models or orchestrate your pipeline; it stores and searches vectors exceptionally well. Features like scalar and binary quantization (shrinking memory footprint several-fold), sparse vector support for hybrid retrieval, and multitenancy via payload partitioning make it a common pick for teams self-hosting serious RAG workloads on a budget.

Day-to-day you’ll run docker run -p 6333:6333 qdrant/qdrant, then use the Python client: client.create_collection(...) with a vector size and distance metric, client.upsert(...) with points and payloads, and client.query_points(...) with a query vector and a Filter of must/should conditions. The REST and gRPC APIs mirror each other, and a built-in web dashboard at port 6333 lets you inspect collections while you develop.