limenn

privacy for AI

protecting your data on Claude, ChatGPT, Gemini and more. names, emails, IBANs and IDs become placeholders before anything leaves your device.

scroll ↓

safely use Claude
without handing over your clients' names and data.

claude.ai

Please write a friendly reminder to Maria Keller (maria.keller@bluewin.ch) that her invoice to IBAN CH93 0076 2011 6238 5295 7 is due on 31.10.2026.

Claude

Dear Maria Keller, a quick reminder that your invoice is due on 31.10.2026. Please transfer the amount to CH93 0076 2011 6238 5295 7. Kind regards

limenn4 values protected · nothing personal left your device

three steps, every time you send

we remove sensitive data

from your prompt and your documents, on your device, before you hit send. names, emails, IBANs and IDs become placeholders like [PERSON-1].

Claude works on the clean request

the AI reads, reasons and writes as usual. it only ever sees the placeholders, never the people and numbers behind them.

your result comes back with the original context

limenn fills your real data back into the answer, on your screen, highlighted, so you can use it right away.

the three steps, up close

one real request, start to finish · click a step

01we remove sensitive datayou type as usual. limenn finds 9 personal values and swaps each one for a placeholder.
02Claude works on the clean requestClaude receives only placeholders and uses them in its answer.
03your result comes backlimenn fills the real values back in on your screen.

limenn bar

the full walk-through, and how to install limenn ↗

what keeps your data safe

"three checks catch what a single filter would miss."

layered detectionformats · names · doubtful cases

"the detection model runs on your own machine."

on your devicemultilingual · in your browser

"contracts and doctor's notes are cleaned before upload."

documents tooPDF + Word · sent as text

"the placeholder vault never leaves this browser."

local vaultcleared when you close it

all features ↗

kinds of personal data recognised

detection stages

MB: a small model that runs on your computer's own processor. no cloud, no graphics card, no special hardware

135

placeholders created for one demo request with a PDF

personal values in outgoing requests to Claude, end-to-end test

see the full evidence ↗

connecting

privacy

intelligence

to make AI safe for sensitive work.

built for people who handle other people's data: medical practices, law firms, HR and finance teams. limenn understands Swiss and European formats out of the box, from AHV numbers and CH IBANs to addresses in Zürich, and works in incognito chats too.

made for sensitive work

medical practices

appointment letters and referrals without patient names

diagnoses can stay, so the AI still understands the case, while the person is pseudonymised…

for teams who can't paste client data into a chatbot

law firms

contract reviews with parties, addresses and IBANs removed

notice periods and clauses go to the AI, the counterparty's identity does not…

HR & finance

employment questions without employee records

salaries, AHV numbers and private phone numbers become placeholders automatically…

all use cases ↗

producteverything limenn does, step by step↗ how it workswhat happens to your data, and how to install↗ evidencetested on 270 Swiss documents, 120 of them written blind↗ researchresults on four public benchmarks, gaps included↗ use casesmedical practices, law firms, HR and finance↗