What happens to your document

1. Upload: the PDF goes to the worker, is processed in memory or temporary files, then deleted. No PDF is retained.
2. Detection: text is analyzed by the Redact model (Modal, GPU). The provider does not persist document content for Veilbrook.
3. Redaction: PDFium/pikepdf; clean pages keep their text; pages with PII are rasterized at 300 dpi with a non-PII text layer.
4. Review: you see items that were not automatically masked and can flag a missed item or “not personal data”. Feedback stores offsets and a decision, not text.
5. Deletion: temporary files are deleted after processing (within 24 hours); metadata is deleted within 30 days after account deletion; metadata backups expire after 14 days. No document content is retained.

What we do not do

• No training on customer content.
• No human access to document content.
• No text logs; logs contain technical events only.
• No sale of personal information.

Subprocessors and email

Modal (GPU inference, United States), Hetzner (application and metadata, European Union), Cloudflare (delivery/DNS), Resend (account emails, European Union) and Stripe when payments are enabled. Details are in the subprocessor list and the DPA. No customer documents are stored on Vast.ai or on the tt-ingest station.

Veilbrook is not SOC 2 certified or HIPAA compliant. A DPA and contractual clauses are available; the public cloud service is being hardened.

Back to the app · What we promise · Terms · Privacy · DPA · Subprocessors