| name | redact |
| description | Strip personal and identifying details out of text so it's safe to paste into a cloud AI, post publicly, or share. Use when the user wants to redact, anonymize, sanitize, or "scrub" something before sending it somewhere else — names, emails, phone numbers, addresses, account numbers, employer, etc. |
| metadata | {"homepage":"https://github.com/bradflaugher/LFG"} |
Redact — scrub it before you share it
This is LFG's job as the privacy front-end to your other AI tools. The user
has a chunk of text — an email thread, a log, a medical note, a contract, a
message — that they want help with from a bigger model (ChatGPT, Gemini,
Claude) or want to post somewhere, but it's full of personal details. You strip
those out on-device so what leaves the phone is safe.
When to use
"Redact this before I paste it into ChatGPT", "anonymize this", "take my name
and address out of this", "make this safe to share".
Instructions
Work entirely from your own reasoning — no tools needed. Everything stays on the
device.
- Find and replace every piece of personally identifying information with a
stable, generic placeholder so the text still makes sense:
- Names of people →
[NAME], [NAME 2], … (reuse the same tag for the same
person so relationships are preserved).
- Organizations / employers / schools →
[ORG].
- Email addresses →
[EMAIL]; phone numbers → [PHONE].
- Street addresses, cities precise enough to locate someone →
[ADDRESS].
- Account numbers, card numbers, IDs, SSNs, policy/case numbers →
[ID].
- Dates of birth, exact ages →
[DOB]. URLs/usernames that identify someone
→ [HANDLE].
- Anything else that could single the person out (rare job title at a named
company, license plate, etc.) → a sensible
[TAG].
- Preserve everything else verbatim — the structure, the meaning, the
non-identifying content. The goal is text a stranger could read without
learning who it's about.
- Output the redacted text first, ready to copy. Then add a short list of
what you removed and the placeholder you used for each, so the user can map
it back themselves if they need to.
Rules
- When unsure whether something is identifying, redact it — err toward privacy.
- Never echo the original sensitive values back except in the brief mapping
list. Don't "summarize" the personal details.
- Don't add commentary about the content itself; just sanitize it.