Qdrant is a vector database and similarity search engine written in Rust. It stores points (vectors with an attached JSON payload) and lets applications search, filter, and manage them, which makes it a retrieval layer for semantic search, retrieval-augmented generation (RAG), recommendations, and agent memory.
It handles dense vectors for semantic similarity, sparse vectors for full-text search, and multivector search for objects with several embeddings or late-interaction models such as ColBERT. Hybrid queries combine several vectors and merge the results with fusion strategies such as Reciprocal Rank Fusion (RRF) and Distribution-Based Score Fusion (DBSF). Payload filters (keyword, full-text, numeric range, geo, and nested conditions) are applied during HNSW graph traversal rather than as a pre- or post-filter.
Qdrant runs as a client-server service with a REST API (OpenAPI 3.0) and a gRPC interface, with official clients for Python, JavaScript and TypeScript, Go, Rust, Java, and .NET. Qdrant Edge is a lightweight in-process variant for edge devices that stores and queries data locally and can sync with a server. Deployment options include self-hosting, the managed Qdrant Cloud on AWS, GCP, or Azure, a hybrid cloud on your own Kubernetes, and air-gapped private cloud.
For coding assistants, Qdrant publishes a set of agent skills that guide decisions such as quantization, sharding, tenant isolation, hybrid search, and model migration.
Features
- Dense, sparse, and multivector search: semantic, full-text, and late-interaction (ColBERT) retrieval in one engine
- Payload filtering: JSON payloads filtered by keyword, full-text, numeric range, geo, and nested conditions with must, should, and must-not clauses
- Hybrid search: multiple vectors per query merged with RRF or DBSF fusion
- Quantization and on-disk storage: built-in quantization cuts RAM use by up to 97% with a tunable speed and precision trade-off
- Distributed deployment: sharding and replication, with collections updated or resized without downtime
- Search tools: faceting, recommendation from positive and negative examples, discovery search, and Maximal Marginal Relevance
- Multitenancy: partitioning of data for multi-user environments
- Hardware acceleration: SIMD on x86-64 and Neon, GPU-accelerated indexing on NVIDIA and AMD, and io_uring async I/O
- Durability: write-ahead logging with update confirmation
- Web UI: browse collections, manage data, call the REST API, and monitor deployment health
- Qdrant Edge: an in-process variant for resource-constrained devices that can sync with a server
- Agent skills: ready-made skills that bring vector search guidance into AI coding assistants
