Qdrant

Qdrant is a vector database for developers building semantic search, retrieval-augmented generation (RAG), recommendations and AI agent memory. It stores embeddings with metadata so applications can find similar items within a relevant subset of their data.

• Hybrid search: Combine semantic and keyword matching with dense and sparse vectors, including BM25, SPLADE++ and miniCOIL.
• Metadata filters: Restrict results by JSON attributes such as text, numeric ranges or location while the search runs.
• Relevance controls: Use multiple embeddings per object for models such as ColBERT, adjust scores with business rules or reduce repetitive results with Maximum Marginal Relevance.
• Storage and scale: Quantization and on-disk storage reduce RAM use, while sharding and replication distribute collections across servers; added vectors become searchable without rebuilding the full index.
• Data inspection: A built-in web interface lets developers explore collections and test queries, alongside REST and gRPC APIs for applications.

The engine is open source under Apache 2.0 and runs locally in Docker or on your servers. It uses CPU acceleration and supports NVIDIA and AMD GPUs for indexing.

Qdrant Cloud is a separate managed service on AWS, GCP or Azure. Cloud Inference generates text and image embeddings there. Hybrid Cloud keeps the data plane on your Kubernetes infrastructure, and Private Cloud supports air-gapped deployments.