qdrant

Qdrant is a high-performance open-source vector similarity search engine written in Rust, designed for production-grade applications requiring fast and reliable nearest neighbor search across large-scale embedding collections. Qdrant emphasizes performance, reliability, and ease of use, leveraging Rust's memory safety and performance characteristics to achieve low-latency queries with minimal resource consumption. Key features include: HNSW indexing with quantization (combining Hierarchical Navigable Small World graphs for fast ANN search with scalar and product quantization for memory efficiency — enabling billion-scale deployments on modest hardware), rich filtering (payload-based filtering during vector search — combining semantic similarity with structured metadata conditions without post-filtering accuracy loss, using a custom filterable HNSW index), multiple distance metrics (cosine, euclidean, dot product, and Manhattan distance), named vectors (storing multiple vector representations per point — e.g., title embedding and content embedding in the same record, enabling different similarity queries on the same data), sparse vectors (supporting both dense and sparse vector representations for hybrid search), collection management (creating, updating, and optimizing collections with configurable parameters), and snapshot and backup capabilities for data durability. Qdrant offers flexible deployment: self-hosted (single node or distributed cluster with Raft consensus for fault tolerance), Qdrant Cloud (managed service), and embedded mode (in-process for development and testing). The architecture uses a segment-based storage system where data is organized into immutable segments with a write-ahead log for durability, enabling consistent performance during concurrent reads and writes. Qdrant's gRPC and REST APIs provide efficient programmatic access, and client libraries are available for Python, JavaScript, Rust, Go, and Java. Qdrant is popular for RAG pipelines, semantic search, recommendation engines, anomaly detection, and image similarity search applications.