Search documents

Find which of your documents likely answer each question, without searching their pages: to choose which to search, or to show a person which documents to read.

found = client.search_documents("What was capital expenditure in 2018?", collection=reports, filter={"year": 2018})

for match in found["results"][0]["documents"]:
    print(match["name"], match["relevance"])

What search_documents() returns, in Python:

{
  "id": "srch_…",
  "results": [
    {"question": "What was capital expenditure in 2018?",
     "documents": [{"id": "pi-cmwi3zsxk000109p6gspztvzb", "name": "3M_2018_10K.pdf", "pageNum": 160,
                    "folderId": null, "collectionId": "cmwi3zsj5000009p64s38ue3v",
                    "metadata": {"ticker": "MMM", "year": 2018}, "relevance": 0.9}],
     "more_documents": […]}
  ],
  "failed_documents": [],
  "processing_documents": [],
  "usage": {"documents": 40, "questions": 1, "billable_documents": 40, "credits": 1.2, "balance": 998.8},
  "seconds": 0.35
}
  • It takes the same scopes as a search of pages, except one document (see Search pages); pass a list of documents as document_ids=, and a folder as folder=folder_id, with recursive=True for its subfolders too.
  • Each question lists the documents likely to help, most relevant first, each with its relevance. It may list none, when no document is likely to: more_documents lists the next most relevant, up to 10, to look at if the question is worded unlike them.
  • top_k: list exactly this many documents instead, the most relevant.
  • A document's file name and metadata count, so name and tag documents well (see Collections with metadata).
  • Each document searched is priced as a document, whatever its length (see Pricing). With metadata, a filter that picks the documents out is free: list them instead.