Skip to main content
Use search when the content is already indexed and you want ranked results immediately.

When search is the right tool

Search is the right fit when you need:
  • low-latency retrieval
  • result ranking for application search
  • grounding material for answer generation
  • collection-scoped access to existing content

Request examples

Node.js and Python examples use the Bulkgrid SDKs. Set BULKGRID_API_KEY in your backend environment; cURL examples also use BULKGRID_BASE_URL=https://bulkgrid.com.

Ranking controls

Bulkgrid currently exposes three main ranking controls:
  • fullTextWeight
  • semanticWeight
  • rrfK
Use the defaults first. Tune only when you have a specific retrieval problem to solve.

Practical tuning guidance

  • increase fullTextWeight when exact keyword matches matter more
  • increase semanticWeight for natural-language search behavior
  • leave rrfK alone unless you are actively testing ranking behavior

Query expansion

queryMode defaults to auto. Set it to none to disable automatic query expansion. You can provide up to three additional strings in queries as query variants. Use the original query for the main user request.

Response handling

Search returns:
  • query
  • scoring
  • results
  • count
Each result includes identifiers, source URL data, chunk text, ranking values, and metadata.

Common product patterns

Search UI

Show:
  • title
  • url
  • trimmed text_content
Keep score available for debugging and ranking analysis.

RAG or answer generation

Pass the top results into your answer-generation step with source URLs attached so the model can be grounded and the UI can cite sources.

Scoped retrieval

Use collectionId to narrow retrieval within the collections allowed by the credential. Enforce user access with scoped credentials and backend authorization; a client-supplied filter is not an access grant.

Failure handling

  • 400: invalid payload or out-of-range values
  • 401: invalid or missing API key
  • 429: back off and retry with jitter