Qdrant

Qdrant is an open source vector database with managed, hybrid and private cloud editions, for teams building RAG, semantic search and agent memory.

Qdrant returns the stored vectors closest to a query vector, filtered by the payload data saved with each point. Applications use it to search by meaning rather than exact words. It is developed by Qdrant Solutions GmbH, an independent Berlin company started in 2021. The engine is open source under Apache 2.0. The same engine runs as a managed service in Qdrant Cloud, inside a customer's network as Hybrid Cloud, or fully isolated as Private Cloud. Points are indexed in an HNSW graph, and payload filters are applied while the graph is searched rather than afterwards. The Query API combines dense vectors with sparse vectors or BM25 full-text matches in one request and fuses the rankings. Scalar, binary and product quantization shrink memory use, while memory-mapped storage keeps vectors on disk. A point can hold several named vectors or a multivector, and prefetched candidates can be rescored or reordered with Maximum Marginal Relevance. On managed clusters, Cloud Inference turns text into vectors at write and query time. Qdrant suits engineering teams that build RAG, semantic or hybrid search, recommendations and agent memory on their own data. Developers who prefer their own infrastructure self-host the engine with Docker, a binary or Helm. Teams that would rather not operate clusters use Qdrant Cloud on AWS, Azure or Google Cloud. SaaS companies keep every tenant in one collection, and regulated companies choose Hybrid or Private Cloud to keep data in their own network. Platform engineers get sharding, shard replication, collection snapshots and JWT tokens limited to chosen collections. Common jobs are retrieval for RAG chatbots, multimodal search across images and text, product recommendations and spotting near-duplicate records. Developers call it from client libraries for Python, TypeScript, Rust, Go, Java and .NET. The FastEmbed library adds cross-encoder rerankers when a team wants them. Infrastructure teams describe their Qdrant Cloud resources as code with Terraform or Pulumi. For an AI agent, a collection serves as long-term memory that takes real-time updates while the agent works.