Semantic search on YDB with YDB-Qdrant

Integration flow

• Choose an embedding model and record its vector dimension.
• Create a collection with matching vectors.size and distance. The sample below uses a three-dimensional toy vector so the curl commands stay copy-pasteable.
• Upsert document chunks or domain records as points with vector and payload.
• Run search/query with a query embedding and a top or limit value.
• Use api-key and optional X-Tenant-Id to isolate environments, tenants, workspaces, or agents.

Create and search

curl -X PUT http://localhost:8080/collections/documents \
  -H 'Content-Type: application/json' \
  -H 'api-key: demo-key' \
  -d '{"vectors":{"size":3,"distance":"Cosine","data_type":"float"}}'

curl -X POST http://localhost:8080/collections/documents/points/upsert \
  -H 'Content-Type: application/json' \
  -H 'api-key: demo-key' \
  -d '{"points":[{"id":"doc-1","vector":[0.1,0.2,0.3],"payload":{"title":"Doc 1"}}]}'

curl -X POST http://localhost:8080/collections/documents/points/search \
  -H 'Content-Type: application/json' \
  -H 'api-key: demo-key' \
  -d '{"vector":[0.1,0.2,0.3],"top":10,"with_payload":true}'

RAG and IDE-agent notes

• Store source ids, titles, paths, or chunk metadata in payloads so search results can be grounded back to source documents.
• Use pathSegments payload filters when the workload needs file-path-oriented search or deletion.
• Use Code Indexer hosted MCP when the target use case is repository memory for coding agents.

When to move beyond this setup

• Move to standalone Qdrant or a managed vector/search platform when full Qdrant parity, ANN indexing, mature hybrid search, or strict low-latency vector SLOs become primary requirements.
• Keep YDB-Qdrant when operational simplicity and YDB co-location matter more than specialized vector database features.