Qdrant
4.5
Qdrant is an open-source vector search engine you can run yourself or buy as a managed service. For teams that cannot send document embeddings to a third party — or that simply want the option of leaving — it is the main alternative to a hosted-only vector database.
- Commercial use OK
- Free tier
- Enterprise contract
- Pricing
- Free
- Free tier
- Free tier
- Commercial use
- Permitted
- Japanese UI
- Not confirmed
- Enterprise contract
- Available
What it is good for
- RAG where documents cannot leave your infrastructure
- Teams that want managed now and self-hosted later, or the reverse
- Workloads with heavy metadata filtering alongside vector search
- Cost-sensitive deployments at sustained volume
Strengths and trade-offs
Strengths
- Open source: self-host with no licence fee
- Managed cloud available with the same engine
- Strong filtering and payload handling
- No lock-in — the same engine runs either way
Trade-offs
- Self-hosting means operating a stateful search cluster
- Smaller managed ecosystem than the hosted-only incumbents
- Tuning for recall and latency is your job
Editors' review
The vector database to choose when you want the option of leaving
Most vector database decisions are made early and are expensive to revisit, because the data is embeddings rather than documents. Qdrant being open source means the exit exists: the managed service and the self-hosted deployment run the same engine.
For organisations that cannot send internal documents to a hosted vendor, that is not a preference, it is the requirement that decides the whole stack.
Running it yourself is real work — a stateful cluster with recall and latency to tune — and the managed ecosystem around it is thinner than the incumbents'. Start managed, keep the self-hosted path open.
Verdict
The right default where data residency matters or where lock-in is the thing you are avoiding.
Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.