Overview

Qdrant is a high-performance vector search engine and database. It helps developers build the retrieval layer of AI applications, storing embeddings and finding the most similar items quickly at any scale and with any deployment model. It is open source and written in Rust, and the project has more than 35,000 GitHub stars and over 60,000 community members. Qdrant says it is SOC 2 and HIPAA compliant.

Use cases

Common uses include retrieval-augmented generation (RAG), recommendation systems, advanced search, data analysis and anomaly detection, and AI agents. Qdrant highlights industries such as ecommerce, legal tech, hospitality and travel, HR tech and healthcare tech.

Search features

You can store metadata as JSON and use advanced filters, including nested, text, geo and has_vector filters. Native hybrid search blends keyword and vector search in one query, supporting dense or sparse vectors with BM25, SPLADE++ and miniCOIL. Built-in multivector support lets you attach several vectors to one object, which makes retrieval more expressive and supports multimodal data. One-stage filtering applies filters during the search itself, rather than before or after, to keep results accurate and fast.

Deployment options

You can run the open-source engine yourself, or use Qdrant Cloud, a managed service with a free starting tier. Qdrant Hybrid Cloud runs in your own infrastructure with Qdrant management, and Qdrant Enterprise Solutions serve larger organisations. Qdrant Cloud Inference creates embeddings inside the platform, Qdrant Edge, in beta, brings vector search to edge devices, and Qdrant Serverless is listed as coming soon.

Developers

Documentation, a roadmap, changelog, certification, benchmarks, demos, a startup programme and a bug bounty programme are available. Guides cover security, observability, quantisation for reducing memory use, and resilience.

Pricing

You can start free in Qdrant Cloud, and the open-source version is free to self-host. Paid cloud pricing depends on resources used; see the pricing page.

Things to consider

Vector search requires choosing an embedding model and tuning chunking and filters for good results. Self-hosting means you handle backups, scaling and updates. Estimate memory needs for large collections before choosing a plan.

Teams often start with the free cloud tier or a local Docker container while prototyping, then move to a managed cluster or their own infrastructure as data and traffic grow, without changing their application code.