Anonymize Text for AI Locally & Restore
Reversible [[P-01]]Anonymize sensitive text for AI in browser RAM—replacing names, emails, phones, and IDs with reversible [[ORG-01]] placeholders before ChatGPT or Claude.
- 100% in browser RAM
- Reversible [[TYPE-NN]] codes
- 0 bytes uploaded
Instant Before & After Preview
Reversible [[ORG-01]] + AI guardrailSubject: Account Review (Ref #A-402)
Contact: maya.chen@example-corp.org · Phone: +1 415-555-0192
Address: 742 Evergreen Terrace, Springfield, IL 62704
Card: 4532-0151-1283-0366 · SSN: 123-45-6789
Please send the follow-up summary to maya.chen@example-corp.org.Subject: Account Review (Ref #A-402)
Contact: [[EMAIL-01]] · Phone: [[TEL-01]]
Address: [[ADDR-01]]
Card: [[CARD-01]] · SSN: [[SSN-01]]
Please send the follow-up summary to [[EMAIL-01]].How it works
How to anonymize text before AI and restore replies
Keep client names, internal projects, and contact details off remote AI servers while preserving full prompt coherence.
- Paste text or highlight custom namesPaste any email, memo, or note; highlight any person or company name in the editor to mask all occurrences globally.
- Copy with AI prompt guardrailClick Copy + AI Guardrail to wrap the masked text in <redacted_document> tags that tell the model to keep [[TYPE-NN]] placeholders intact.
- Restore the AI reply locallySwitch to the Restore tab and paste the AI response—your browser RAM vault swaps every placeholder back and flags any mangled or missing tags.
FAQ
Frequently asked questions
Why does TxtTo use [[ORG-01]] double brackets instead of [ORG_1] or fake names?
Double uppercase brackets without underscores avoid Markdown italic mangling, prevent cross-language translation by LLMs, and remain deterministic for exact restoration.
Where is the mapping between original text and [[P-01]] stored?
Only inside current-tab browser RAM, clearing automatically on refresh. If needed, you can export an AES-256-GCM passphrase-encrypted .redactkey file locally.
What happens if the AI accidentally drops a bracket or changes case like [org 1]?
The 3-tier Restore engine fuzzy-repairs mild formatting drift in amber so you can review it immediately, while flagging hallucinated or missing codes in red.