cookbook / qdrant
Qdrant
A vector database, for a knowledge file the agent can search by meaning, not grep.
- database
- vector
- ai
Run it
briq_run({
recipe: "qdrant",
params: {
"size": "small"
},
wait: true
})briq run qdrant
briq ls
briq logs -f <id>
briq rm <id>import { Briq } from '@briq/sdk';
const briq = new Briq(process.env.BRIQ_API_KEY);
const qdrant = await briq.run({
recipe: 'qdrant',
params: {
"size": "small"
},
});
console.log(qdrant.outputs.QDRANT_URL);
// ... when done
await briq.destroy(qdrant.id);Parameters
| Name | Type | Options | Default |
|---|---|---|---|
| size | enum | nano, small, medium, large | "small" |
Outputs
Returned by briq_run once the stack is healthy. Hosts resolve inside your team network; secrets are generated per run.
- QDRANT_URL
- http://qdrant.<stack>.internal:6333
- QDRANT_GRPC_URL
- http://qdrant.<stack>.internal:6334
- QDRANT_API_KEY
- <QDRANT_API_KEY>
What it runs
| Briq | Image | Ports | Size | Volume |
|---|---|---|---|---|
| qdrant | qdrant/qdrant:latest | 6333, 6334 | small | 10 GB at /qdrant/storage |
When to use it
- The agent's context is a growing pile of documents, transcripts or past runs, and grepping a markdown file for a keyword misses everything phrased differently.
- You are building or testing retrieval-augmented generation and need a real vector index — upsert, filter, and nearest-neighbour search — not a mock.
- You want the agent to embed what it learns as it goes and query it back by meaning in the same session, instead of re-reading and re-summarizing a note file each time.
The volume is 10 GB, mounted at /qdrant/storage, so a stopped briq keeps its collections.
Tips
- The API key is required, not optional. This recipe sets it via
QDRANT__SERVICE__API_KEY; pass it as a header (api-key: <QDRANT_API_KEY>) on every request. - Use gRPC for bulk upserts.
QDRANT_GRPC_URLis materially faster than REST for inserting many points at once; useQDRANT_URLfor the dashboard and one-off queries. - The dashboard is worth showing a human.
briq_exposeport 6333 withmode: "token"and open/dashboardon the resulting URL to browse collections visually. - Collections aren't created for you. The agent creates them on first use with a
PUTto/collections/<name>naming the vector size and distance metric.