Trust & privacy
Your document.
Processed, then discarded.
We process your document in memory, return the redacted copy and discard both. We don't store document content or use it to train models.
Try it with an exampleFrom your device. Back to you.
Send it securely
Your text or PDF travels over an encrypted connection.
Remove personal details
Our own infrastructure finds and removes identifying information. Your content isn't sent to a third-party AI service.
Review your copy
Download your result, then check it before sharing. We discard the original and the redacted copy after processing.
Measured, not assumed
How well does it work?
Our CV benchmark checks whether personal information is fully removed. An item passes only when every letter and digit is hidden.
The test uses generated CV text and default instructions. Results on your documents or with custom instructions can differ, so always review the output.
99.7%
of labelled PII items fully redacted
958 labelled items across 60 synthetic CVs, using default instructions.
A text benchmark, not a guarantee that a real document is anonymous.
How the benchmark is measured
99.7% of 958 labelled PII items were fully redacted across 60 synthetic CVs in our CV benchmark. A name or other labelled item counts as redacted only if every letter and digit is hidden. This measures individual items, not how many documents are guaranteed anonymous.
Separate PDF tests use generated one- and two-column CVs and read the resulting pages to check for information still visible. The 99.7% text result is not a score for those PDF checks.
More than looking for a pattern
Find it. Match it. Remove it.
Personal details don't always look like an email address or phone number. Names, employers and schools need context too.
Understand the context
AI reads the text to find identifying details. Account instructions let you change what it looks for.
Find the matches
Matching rules locate those details across the document, including repeated mentions, line breaks and spacing differences.
Build a clean copy
Text gets replacement labels. PDFs get masked areas and rebuilt pages, without the original text layer or attachments.
What changes in a redacted PDF?
Detected areas and embedded images are masked, then every page is rebuilt as an image. The original text layer, metadata, bookmarks, forms and attachments are left out. Text under a redaction bar cannot be recovered from that output.
The result is no longer searchable or selectable text. Information that wasn't detected may still be visible. PDFs need selectable text to begin with; scanned and image-only pages aren't supported.
What if redaction fails?
We return an error instead of unchecked content. If any section fails detection, the whole request fails. Failed redactions aren't charged.
What information do you keep?
We keep usage records for billing, support and preventing abuse: the account, time, size, page count, number of redactions, duration and outcome. These records are normally retained for one year; usage not yet billed is kept until it has been reported. They don't contain your text, file name, detected details or redacted copy.
The free daily allowance uses a one-way hash of your IP address that changes each day. Those counts are deleted after two days. Account information and saved redaction instructions follow our privacy policy.
What about payments and connected AI tools?
Stripe handles payments; we don't see or store your card number. When you use an AI assistant, it decides which content to send us. Your assistant provider handles the conversation and attachments under its own policies. Redacting a copy doesn't erase content already shared with that provider.
The people behind the product
Built by Hexploits.
We built PII Purge to help people share useful documents with less personal information in them. Use it for a one-off job, connect it to Claude or bring it into your own application.
PII Purge is operated by Hexploits Ltd, registered in England and Wales.
- Company number
- 11669296
- Registered office
- 71–75 Shelton Street, Covent Garden, London WC2H 9JQ
- VAT number
- GB481356383
- ICO registration
- ZB895216
A few helpful answers
Frequently asked questions
What makes PII Purge different from a standalone PII analyser?
Our multi-step pipeline combines contextual detection, matching and document processing. Detection finds identifying details in context; matching rules locate repeated mentions and text split across lines; PDF processing maps those matches to the page, masks them and rebuilds the file without its original text layer. These stages form one redaction service, with a result you can review before sharing.
Does the CV benchmark guarantee an anonymous document?
No. The benchmark measures the share of labelled personal-information items fully redacted, not the share of CVs guaranteed to be anonymous. It uses synthetic CV text and default instructions. Your document, its layout and custom instructions can affect the result. Always review every page before sharing.
How are detection and PDF removal different?
Detection identifies the information to redact. Document matching locates it in the original text and maps it to positions on PDF pages. The PDF is then rebuilt from redacted page images, without its original text layer, metadata or attachments. Information that was not detected may remain visible, so review is still required.
What happens if detection cannot complete?
The service returns an error instead of unchecked content. If a document is split into sections for detection and any section fails, the whole request fails. Scanned pages without a text layer are rejected because OCR is not supported.
How accurate is PII Purge on CVs?
99.7% of 958 labelled PII items were fully redacted across 60 synthetic CVs in our CV benchmark. The test uses synthetic CV text and default instructions. An item counts as fully redacted only when every letter and digit is hidden. Results can differ on your documents or with custom instructions, so always review the output before sharing.