Connect Qdrant

Qdrant runs as an external vector database. Use it when you want to separate search data from the SpaceBuilder server or operate a larger vector set. Source files remain in SpaceBuilder; Qdrant stores vectors and the payload needed for retrieval.

Before you begin

• Required role: Workspace administrator and access to the Qdrant connection information
• The SpaceBuilder server must reach Qdrant's REST and gRPC endpoints.
• Confirm the output dimension of the embedding model.
• If no Qdrant service is available, follow Set up a Qdrant server.

Qdrant's official local quickstart documents REST on port 6333, gRPC on 6334, and the dashboard under the REST endpoint. Qdrant starts without authentication by default, so restrict network access or enable an API key and TLS for a shared environment.

Configure the connection

1. Open Knowledge Services > Knowledge Storage.
2. Select Qdrant as the driver.
3. Enter the connection values.
4. Select Save.

If the gRPC endpoint is empty and the REST endpoint uses port 6333, SetFN derives port 6334. It falls back to REST when gRPC is unavailable.

Let SetFN create the collection

Qdrant collections are created on the first vector write. SetFN creates a single unnamed dense vector with the configured dimension and Cosine distance, then adds payload indexes used for filtering.

Before the first indexing run, the status can report that the collection will be created later. This is an expected state.

If you must create the collection manually, use the REST request below and set size to the embedding dimension. This layout matches Qdrant's official collection API.

PUT /collections/setfn_knowledge_example
{
  "vectors": { "size": 1024, "distance": "Cosine" }
}

Do not configure a named vector or hybrid layout for this collection. SetFN expects a single unnamed dense vector. A different dimension or distance metric is reported as incompatible.

Verify stored data

Knowledge Storage reports connection state, this installation's entity count, total collection count, and collection compatibility. One vector normally represents one document chunk, so vector count is usually higher than document count.

Use the Qdrant dashboard at http://<host>:6333/dashboard to inspect the service. You can inspect sample payloads without retrieving vectors:

POST /collections/setfn_knowledge_example/points/scroll
{ "limit": 10, "with_payload": ["source_path", "text"], "with_vector": false }

For Qdrant 1.12 or later, count values by source path with the facet API:

POST /collections/setfn_knowledge_example/facet
{ "key": "source_path", "limit": 50 }

Share a Qdrant service

SetFN includes an installation isolation key in new vectors and filters searches and deletion to that key. Separate collections are still recommended for simpler capacity management, backup, and removal. The settings page warns when another local SpaceBuilder project uses the same collection.

Related documentation

• Knowledge Storage
• Operating Knowledge Storage
• Set up a Qdrant server
• Connect Milvus
• Ingestion & Indexing