Plain-English trust notes

How DocRinse handles documents.

DocRinse is built around a simple idea: reduce exposure before AI. Pasted text now gets a browser-side first pass for obvious identifiers. Full file workflows still use the DocRinse backend to create a cleaned version, return it to the browser for review, then let the user decide whether to export the cleaned output.

Local-first privacy direction

Pasted text preflight

For pasted-text runs, obvious SSNs, emails, phone numbers, address-like lines, birth-date labels, and account-style labels are replaced in the browser before backend review.

Backend review still matters

The browser pass is intentionally simple. The backend still performs the stronger document-type detection, privacy setting, redaction report, and cleaned-result generation.

File workflows

PDF, DOCX, TXT, and OCR workflows are not fully local today. Uploaded files are sent to the DocRinse backend for that review run.

The product direction is local-first cleaning where it is practical. The current honest claim is narrower: pasted text gets browser-side preflight before server review; full local PDF/OCR processing remains roadmap work.

Current processing model

Original input

Your uploaded file or browser-preflighted pasted text is used to extract text, detect private details, and create a cleaned version for that run.

Browser-held review copy

The cleaned output is returned to your browser for review/export. The backend keeps temporary run metadata, not a plaintext cleaned-document cache.

Not a document vault

DocRinse is not intended for long-term document storage. Anything you want to keep should be exported from the cleaned output.

The original document content is not recorded as long-term product history. Audit records use hashes, counts, categories, and processing metadata instead of raw private values.

What happens in one run

1. Submit

You upload a supported file or paste text. Pasted text gets the browser preflight first; uploaded files are sent over HTTPS for processing.

2. Extract

DocRinse reads text from the file or form input. Text-based files are the most reliable lane; OCR is used only when needed and carries stricter review warnings.

3. Detect

The backend classifies the document and flags likely private details such as names, account data, contact details, dates, IDs, and document-specific fields.

4. Clean

Private-looking values are replaced with placeholders, while the system tries to preserve useful document context, labels, amounts, and structure.

5. Review

The cleaned result and redaction report return to the browser. Export controls stay gated until the user marks the result reviewed.

6. Export or discard

Paid exports are built from the browser-held cleaned result. Temporary backend metadata expires, and the user can discard the run sooner from the app.

This workflow is intentionally a review system, not an autopilot. If a document is scanned, unusually formatted, legally sensitive, medical, or compliance-heavy, treat the cleaned output as a draft that still needs human inspection.

Where uploads go

Upload destination

Full document workflows upload to the DocRinse processing app at api.docrinse.com over HTTPS. Pasted-text preview sends the browser-preflighted text to the same processing app. The public marketing site is served separately from www.docrinse.com.

Cloud hosting

The public site runs on Netlify. The processing backend runs on Railway. Paid checkout is handled by Stripe.

Operator

DocRinse is operated by Kingly Industries, a DBA of Kingly Insights LLC. The upload target is a DocRinse-controlled service, not an undisclosed personal machine.

Data residency should be treated as a procurement question, not a slogan. The current public product is transparent about its hosting stack, but it is not yet sold as an enterprise data-residency guarantee. Formal buyer reviews should verify the active Railway deployment region and subprocessors against the current service configuration.

How a buyer can verify the path

Current validation evidence

DocRinse trust claims are backed by repeatable internal QA gates. These are not customer-facing accuracy guarantees, and users should still review the cleaned output before export.

9/9

public-domain prose excerpts passed the preservation loop without private-detail false positives.

7/7

external public and mock PDFs passed with no watched private terms leaking after rinse.

34/34

synthetic adversarial privacy documents passed with zero watched answer-key term leaks.

The practical meaning is simple: DocRinse is being tested against both over-redaction risk and privacy-leak risk before public trust copy moves forward. The user still owns final review.

Using the cleaned result with AI

DocRinse separates document cleaning from whatever AI tool you use afterward. The cleaned version is reviewed first, then you decide where to take it.

DocRinse does not run AI inside the product. It produces a cleaned result for you to review and export.

Feedback and payments

Run feedback

If you submit feedback, DocRinse stores the feedback type, optional note, document classification, OCR status, and redaction summary. Do not include raw personal details in feedback notes.

Stripe checkout

Paid checkout is handled by Stripe. DocRinse receives a verified payment session and issues a temporary access code for the selected plan.

Rate and file limits

Production limits apply to request rate, file size, pasted-text length, page count, and paid-plan capabilities.

Feedback notes are sanitized before storage, but the safest path is still simple: do not paste private values into feedback or support notes.

Important limits

DocRinse helps identify sensitive information for review and creates a cleaned version. It does not guarantee that every possible sensitive detail will be found. You should inspect the cleaned result before exporting it or using it with AI.