What happens to your documents

A document you upload is used for one thing: running your review. We extract its text, structure it into spec requirements or submittal claims, and compare the two.

The product only ever shows you your own reviews. Every request for a review, or for anything extracted from it, is checked against your account before any record is returned.

The PDF files themselves are stored in a private cloud bucket, not a public one. They are never given a public URL. Every link to a file is a signed link, generated under your session and expiring within an hour. Anyone you forward that link to before it expires can open it, so treat it the way you’d treat the document itself.

• OpenAI — structures spec text, extracts submittal claims, and compares them, through OpenAI’s API (not the consumer ChatGPT app). Data submitted through the API is not used to train OpenAI’s models; see OpenAI’s data-usage policy. OpenAI may retain API inputs and outputs for a limited period for abuse monitoring, so deleting a review here does not necessarily erase OpenAI’s own copy on the same day.
• AWS Textract — reads the text and layout off each PDF page (OCR) so the content can be structured and compared. Under the AWS Service Terms, AWS may retain content processed by its AI services for a limited period unless the account has an AI services opt-out policy in place.
• Supabase — hosts the database and the file storage that hold your documents and review data, on AWS infrastructure.

Your account has a retention setting — 7, 30 (the default), or 90 days — that you can change at any time in Settings.

To be straight with you about what that setting does today: automatic deletion isn’t running yet, so the setting doesn’t delete anything on its own. Uploads, and everything extracted from them, are kept until you delete the review. We say the same thing next to the setting itself.

You can delete any individual review immediately, whenever you want, regardless of that setting — this really does remove the PDFs and everything extracted from them, right away (see below).

Deleting a review removes:

• the uploaded PDF files (the spec and the submittal),
• the text our OCR extracted from them,
• the spec requirements we identified,
• the submittal claims we identified,
• every comparison result generated from them, and
• the rest of the review’s history — job records, manual labels, and anything else our pipeline produced along the way.

It is not a soft delete of a database row — inside Arc Tech, the underlying content goes with it.

To be exact about the boundary: that covers Arc Tech’s own database and file storage. The processors listed above run on their own retention schedules, which we do not control — see what each of them says under “Who processes them.”

• We do not use your documents to train OpenAI’s models — see the policy linked above.
• We do not share your documents, extracted data, or results with any other account or firm. We don’t build features that expose one account’s reviews to another, and every request is checked against the account that made it, as described above.
• We never generate a public link to your documents. Every link we create is signed, generated under your session, and expires within an hour.

Email support@arctechai.com — a real inbox, answered by a person.