Skip to main content
Use Semantic Search when you want the assistant to retrieve information from previously ingested artifacts (documents, notes, PDFs, HTML, and more).

​
Prerequisites

​
Requirements

  • Tool spec: { "name": "semantic_search", "type": "semantic_search_v1" }
  • tool_context: at least one ingested_artifact with a context_filter

​
Basic request and response

This example validates a request that enables semantic search over a collection: 
curl -X POST "https://{BASE_URL}/api/gpt/v1/messages/validate" \
  -H "Authorization: Bearer {API_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "default",
    "max_tokens": 128,
    "messages": [
      { "role": "user", "content": "What shipped in the latest billing dashboard release?" }
    ],
    "tools": [
      { "name": "semantic_search", "type": "semantic_search_v1" }
    ],
    "tool_choice": { "type": "auto" },
    "tool_context": [
      {
        "type": "ingested_artifact",
        "context_filter": {
          "collection": "{collectionID}"
        }
      }
    ]
  }'
Example response: 
{ "valid": true, "errors": null }
To run the request, send the same body to POST /messages.

​
Tips

  • Keep tool_context narrow (use a single collection and the smallest set of artifacts you need).
  • If you’re not getting relevant results, ingest cleaner source text (or split large documents into smaller artifacts).

​
Next steps

  • Ingest and manage artifacts: Artifacts
  • Learn how messages return tool blocks: Messages