Skip to main content

Claims

Extract atomic, importance-scored factual statements from each document so they can be reasoned about programmatically.

Extract claims from a collection#

Run claim extraction and stream the resulting claim map.

Claims are atomic factual statements pulled from each document section. The map-stream runs the extractor end-to-end and yields progress events; subscribe and update a UI as it goes. A collection has to have `enableSummarization` and a provider key (or a project-level shared LLM model) configured before extraction can run.

dewey claims list doc_a1b2c3 --json | jq '.[] | {section: .sectionTitle, importance, text}'

See also

List claims for a document#

Get importance-scored facts; filter by importance or section.

Each claim carries an `importance` from 1 (incidental) to 5 (load-bearing), the source `sectionTitle`, and the chunk text it was drawn from. Filter to `minImportance: 3` for headline-only summaries; pull everything for full-fidelity downstream use.

dewey claims list doc_a1b2c3 --min-importance 3

Response

{
  "documentId": "doc_a1",
  "filename": "report.pdf",
  "claims": [
    {
      "id": "clm_x1",
      "sectionTitle": "Revenue",
      "sectionLineage": ["Executive summary", "Revenue"],
      "text": "Net revenue grew 18% year-over-year in Q3 2026.",
      "importance": 5,
      "position": 0
    }
  ]
}