PDF text evidence API
Extract bounded, page-cited text snippets from a verified PDF for document automation and grounded AI retrieval.
Cited text without hallucinations#
Use the document evidence endpoint after uploading and completing a PDF. PDFHQ validates the exact SHA-256 source revision, extracts embedded text from each page, and returns bounded, page-labeled snippets with offsets. This is a deterministic extraction utility, NOT a completed OCR or LLM service.
curl 'https://api.pdfhq.io/v1/documents/uploads/UPLOAD_ID/evidence' \
-H 'x-api-key: YOUR_API_KEY'Source revisions bind citations to the uploaded bytes; offsets refer to normalized page text, not to pixel-coordinate highlights. The API limits documents to 12 MiB and 40 pages, with bounded text output. Image-only scanned pages return no text without OCR.
Privacy and quota safeguards#
Only project API keys with enabled API entitlements can use evidence extraction. The central HQTools monthly API call quota is consumed before fetching sealed document bytes; anonymous and browser-only access is rejected. The response is private, no-store, and document text is not persisted as embeddings or sent to a model by this endpoint.
Do not trust instructions embedded inside extracted PDFs. Verify citation source_revision and page/offset values before using them as evidence in AI answer generation.
