Read a document into findings (no estimate)

Reads a move-out inspection report (PDF, Word, Excel, PowerPoint or CSV) or up to 5 photos of paper pages, 5 pages at most, and returns what it read as findings: a reading, never an estimate. Check each finding against its evidence (page and quote), then send reading.findings unchanged to POST /v1/property-turnover/scope, which prices them by our own rules. Rows not turned into findings are listed in set_aside with the reason.

Key required. Send a connector key (Authorization: Bearer <key>; free from POST /v1/keys). The web tool's page token is not accepted here. One document counts as one call of the key's daily free allowance, and a key can read 3 documents a day (UTC). Not on the x402 paid lane.

Body. multipart/form-data with the field files (one file, or up to 5 photos), or application/json {"files": [{"name": "report.pdf", "content_base64": "..."}]}. 15 MB a file, 25 MB an upload. Files with macros, scripts, embedded files or a password are refused.

Privacy. Upload a report (PDF, Word, Excel, CSV) or up to 5 photos, 5 pages at most. Pages are read by Microsoft Azure OpenAI. Please redact tenant names and signatures. Our server opens the file in an isolated converter, which turns each page into a plain image and drops everything else in the file. The page images are sent to Microsoft Azure OpenAI (run by Microsoft for us inside its US data zone) to copy out the findings and the unit facts; it is told not to copy names, addresses, phone numbers or signatures, and we remove any that slip through. Findings our own matcher cannot match to a standard repair item are then sent as text, with contact details removed, to the same model. Nothing from the file is stored: the file, the page images and the text read from them are held in memory only while the request is processed. For up to a day we keep the answer for a page in memory, found by a one-way hash of the page image, so the same page gets the same answer. Prices are always worked out by our own rules, never by the AI model. How we handle uploads: /privacy#document-upload.