Set up Qdrant
AdminQdrant is the most powerful vector backend for semantic search in the knowledge base. When Qdrant is worth using over sqlite-vec or BM25 is explained in RAG & search.
Where is it configured?
Section titled “Where is it configured?”You maintain the connection in the Qdrant section of the System settings:
- Endpoint (URL of the Qdrant server),
- API key (stored encrypted),
- Collection name, vector dimension, search limit, minimum score.
A test connection saves the values and checks reachability.
Alternatively, these values can be set via appsettings/environment variables (keys Qdrant:VectorCollectionName, Qdrant:VectorDimension, Qdrant:SearchLimit, Qdrant:MinimumScore, Qdrant:AspireConnectionName). → Environment variables & appsettings
Backend selection
Section titled “Backend selection”Which backend is active is controlled by Knowledge:VectorBackend:
- Auto (default) – uses Qdrant if configured; otherwise sqlite-vec (if embeddings are available); otherwise BM25 as a fallback.
- Qdrant / SqliteVec – fixed choice.
In desktop mode, the backend can additionally be switched via a selection in the System settings; in web/API mode it is resolved at startup.
Provisioning
Section titled “Provisioning”- In the web stack, Qdrant can be co-orchestrated via the Aspire AppHost (connection name
qdrant). → Multi-user – API self-hosted - Single workstation with Docker describes the local setup. → Single workstation with Docker
- A ready-made, best-practice Compose setup is provided in Self-host Qdrant (Docker Compose).
Limits & notes
Section titled “Limits & notes”- Without Qdrant and without sqlite-vec, the desktop app uses BM25 (lower result quality for semantic/German search queries).
- The API key is never shown in plain text.
Related
Section titled “Related”- Self-host Qdrant (Docker Compose)
- RAG & search · Admin tab: Knowledge areas & indexing
- Embedding models across workstations & vaults (several workstations, one Qdrant?)
- Set up Presidio (analogous external service)