Skip to main content
Semantic search over the workspace document library. Display name “Retrieve Documents”. On by default for new agents. Each hit includes a file_url you can pass straight to structured_data_extraction.

​
Authentication and enablement

No extra integration. Searches documents already ingested for the workspace. Default-on.

​
Inputs

  • query (required): question, topic, or description. Max 10,000 characters.
  • limit (optional): 1–20, default 5.

​
Output

Matching documents with text, metadata, extracted data, and file_url.

​
Limits and side effects

  • Vector similarity only — not keyword SQL and not live web search (web_search).
  • Read-only against the library.

​
Expected errors

  • Empty or oversized query.
  • Embedding / search backend failure.