Qdrant
github.com/qdrant/qdrant- Category
- Developer Tools
- Rank
- No. 339Tools index
- Pricing
- Freemium
- Type
- TOOL
- Builder
- qdrant
- GitHub
- 34.0k stars
- Latest release
- v1.19.0
- Added
- Jul 4, 2026
About
Qdrant is a high-performance, open-source vector database and similarity-search engine written in Rust, built for storing and searching embeddings at massive scale for RAG and AI applications.
What it does
Qdrant organizes each record as a vector plus JSON payload. Queries compare vectors, apply payload conditions, and can combine semantic and keyword signals through configurable fusion. It handles dense, sparse, and multivector data, while supporting faceting, recommendations, discovery, and relevance tuning.
Why it's ranked here
The case for Qdrant is strong because it covers both search quality and production operations. Hybrid retrieval, rich filtering, quantization, sharding, replication, persistence, metrics, telemetry, and audit logging form a coherent system. REST, gRPC, official clients, a web interface, and an in-process edge option make adoption practical across varied stacks.
What's good
Filtering is unusually capable: JSON payload conditions include keywords, full text, numeric ranges, and locations, with composable Boolean clauses. Hybrid queries merge several vectors using configurable fusion strategies. Quantization offers an explicit memory, speed, and precision tradeoff. Horizontal scaling includes sharding, replication, and collection resizing without downtime. Write-ahead logging protects confirmed updates, while payload indexes inform query planning.
Tradeoffs
The quick local container starts without authentication and listens on every network interface, so it is unsafe until configured. Production deployment requires attention to security, installation, storage compatibility, sharding, replication, and tuning. Some acceleration is conditional: GPU indexing needs supported hardware, while the asynchronous disk backend is Linux-specific and other systems fall back to memory mapping.
How to use it well
Use Qdrant when an application needs semantic, keyword, or hybrid retrieval with metadata filters, recommendations, or faceting. Start with the container and an official client, then secure the service before exposing it. Choose the edge edition for local, offline, low-latency access with later server synchronization. Bring your own embedding model and tokenizer pipeline; Qdrant handles retrieval and storage, not general model inference.
Technical notes+
Cargo.toml defines a Rust 2024 workspace split across API, collection, segment, shard, storage, WAL, GPU, BM25, and Edge crates. src/actix/mod.rs builds the REST service with Actix, optional API-key authentication, TLS, validation, telemetry, health routes, and the web UI. src/tonic/mod.rs exposes public gRPC services, reflection, gzip, optional TLS and authentication, plus internal cluster services; it selects io_uring storage on Linux and mmap elsewhere. src/main.rs starts a peer/server, handles bootstrap, snapshots, telemetry controls, consensus recovery, filesystem checks, and optional GPU initialization. lib/wal/src/lib.rs implements segmented, locked write-ahead storage and recovery checks. lib/bm25/src/lib.rs provides compute-only sparse embeddings over caller-supplied tokens. lib/edge/src/lib.rs exposes in-process shard, query, update, and read-only APIs.
Observed
- License
- Apache License 2.0
- Primary language
- Rust, using the 2024 edition
- Install surface
- A Docker image runs the server locally; a managed cloud service and an in-process Edge edition are also offered.
- Network interfaces
- REST with an OpenAPI 3.0 specification and gRPC are supported.
- Client libraries
- Official clients are listed for Go, Rust, JavaScript and TypeScript, Python, .NET and C#, and Java.
- Application interfaces
- The repository describes a server CLI, web UI, REST API, gRPC API, and Python or Rust access to the Edge edition.
- Hardware and operating-system paths
- Indexing can use NVIDIA or AMD GPUs; Linux uses an io_uring storage backend while other systems fall back to memory mapping.
Read from README.md, Cargo.toml, src/main.rs, src/actix/mod.rs, src/tonic/mod.rs, src/common/mod.rs, src/tracing/mod.rs, src/migrations/mod.rs, lib/api/src/lib.rs, lib/gpu/src/lib.rs, lib/wal/src/lib.rs, lib/bm25/src/lib.rs, lib/edge/src/lib.rs.
Tags
Tech Stack
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.