Qdrant

Rust-fast open-source vector engine. Cleaner API than Weaviate, smaller footprint.

VISIT QDRANT

Key takeaway: Filtered search is the part of RAG that quietly falls over, and Qdrant is the engine built around getting it right. Reach for it when every query carries a tenant, a user, or a date range alongside the vector. Weaviate is the better answer if built-in keyword fusion or vectorizer modules matter more to you.

Quick facts

  • CategorySelf-hosted
  • EngineRust
  • PricingFreemium
  • LicenseApache-2.0
  • Created2020
  • GitHub stars28.4k
  • Hybrid searchNative
  • Edge-readyNo
  • Multi-tenantNative
  • Max dimensions65,535

What it is

Qdrant is a Rust-based open-source vector database. Cleaner API and simpler operational model than Weaviate, strong filter performance (filterable HNSW indexes are a real differentiator), Qdrant Cloud for managed hosting. Used heavily in production AI deployments where latency at scale matters.

Further reading

Best for

  • Latency-critical RAG workloads
  • Apps with heavy filtered search (filter by user_id, tenant, time, then vector-rank)
  • Self-host-first teams who value Rust's performance and operational simplicity

When not to pick it

Skip Qdrant if hybrid search (vector + keyword) is the deciding feature, Weaviate is stronger there. Skip for very small workloads where pgvector suffices.

My take

Qdrant is the right pick for self-hosted vector at scale when filtering is the hot path. Rust performance is real; the API is cleaner than the alternatives.

Why filtering is the differentiator

Most engines treat a filter as something applied before or after the approximate search, and both orders have failure modes. Pre-filter and you may scan far too much; post-filter and a restrictive condition can return almost nothing, because the candidate set was chosen without knowing about it. Qdrant builds the filter into the graph traversal itself, so a query constrained to one tenant or one date window still explores the right region of the index. In practice that is what stops a multi-tenant RAG app returning three results when it should return twenty. Payload indexes on the fields you filter by are the thing to get right and the thing teams forget. Quantization options let you trade a little recall for a large drop in memory, which is usually the lever that keeps a self-hosted deployment on sensibly sized machines.

Running it, and where it stops

Qdrant is a single Rust binary with a small footprint, which makes it one of the least annoying database services to operate. It starts fast, it does not drag a message queue and a coordination service along behind it, and the REST and gRPC APIs are clean enough that you rarely reach for the documentation twice. Distributed sharding and replication exist but are younger and less battle-tested than Milvus's, so a genuinely billion-scale deployment is not where Qdrant is strongest. The server also will not embed for you the way Weaviate's vectorizer modules do, so that step is yours to build: more flexibility, more code. Hybrid search works through sparse vectors and result fusion, which is capable but less turnkey than Weaviate's BM25 path. Qdrant Cloud is there when you want the engine without the deployment work.

Frequently asked questions

Is Qdrant free?

Yes. The engine is open source under Apache-2.0 with no crippled community edition, so self-hosting costs only the machines you run it on and the time to keep them healthy. Qdrant Cloud is the paid managed service and has a free tier for small workloads. Most teams start on that free tier and make the self-hosting decision later, once traffic is real.

Qdrant or pgvector?

pgvector if Postgres is already in the stack and the workload fits comfortably inside a single instance, because one database beats two every time. Qdrant once vector search starts competing with your transactional workload, once filtered queries need to stay fast at scale, or once you want quantization and horizontal scaling that Postgres will not give you.

Does Qdrant do hybrid search?

Yes, through sparse vectors alongside dense ones and a fusion step that combines the two rankings. It works well, but it is more assembly than Weaviate, where BM25 is built into the engine and a single parameter balances the two signals. If hybrid retrieval is the deciding requirement rather than a nice-to-have, evaluate Weaviate alongside it before committing.

Links

Compare Qdrant side-by-side

Qdrant vs Weaviate → Qdrant vs Pinecone → Qdrant vs pgvector →

Similar tools you should also consider

Weaviate

Open-source self-hostable vector database with hybrid search and module ecosystem.

Read the take →

Milvus

Distributed vector DB built for billion-scale workloads. Heavy, capable, China-AI-aligned.

Read the take →

pgvector

Vector search inside Postgres. The default for teams already on Postgres or Supabase.

Read the take →

If Qdrant is your pick, the next conversation is short

The 30-min call is where your vector-DB choice becomes a real RAG architecture, a chunking + reranking strategy that actually works for your corpus, and a price range you can take to your stakeholders. Describe your data shape, your query patterns, your latency budget. I tell you whether Qdrant is genuinely your fit.