hemlis/Docs/How it works
v1.0 · updated May 8Back to app
How it works

Anonymize data before
sharing with AI.

Hemlis replaces sensitive values in spreadsheets and documents with anonymous placeholders — so you can share data safely with AI agents without exposing real names, companies, or identifiers. Decode the response to restore originals.

Client-side onlyExcel · CSV · Word · PDF · Text
01
Upload
Excel, CSV, Word, PDF, text
02
Anonymize
Column prefixes + keywords
03
Export
Anon file + decoding key
04
Share
Get AI insights, safely
05
Decode
Restore originals locally
01 / ANONYMIZE

Configure rules

Column rules

Pick a column, set a prefix. Each unique value gets a sequential ID.

CustomerCustomer · anonymized
Acme CorpCUSTOMER_001
Globex IncCUSTOMER_002
InitechCUSTOMER_003

Keyword rules

Global find & replace that works on any file type.

John Smith→PERSON_A
Project Phoenix→PROJECT_X
Stockholm HQ→LOCATION_1
02 / PREVIEW & EXPORT

Review and export

Toggle Original / Anonymized to verify in place. When you're ready, the export bar gives you three artefacts: the cleaned file, the decoding key, and a reusable rules preset.

Anonymized data
sales.anon.xlsx
Decoding key
sales.key.xlsx
Sensitive — store securely
Rules preset
sales.rules.xlsx
03 / SHARE

Send it to the AI

you ›
Analyze this sales data and flag customers at risk of churning.
ai ›
Based on the data, CUSTOMER_002 shows a 23% decline in order frequency. CUSTOMER_005 has reduced their average order value by 40%. I'd flag both as churn risks and recommend outreach from PERSON_A's team.
04 / DECODE

Restore originals locally

Switch to the Decode tab, paste the AI's reply, and the tokens get swapped back to real values — without the mapping ever leaving your browser.

Decoded output
Based on the data, Globex Inc shows a 23% decline in order frequency. Wayne Enterprises has reduced their average order value by 40%. Recommend outreach from John Smith's team.
REFERENCE

File naming convention

.anon.*sales.anon.xlsxAnonymized data — safe to share with an AIsafe to share
.key.xlsxsales.key.xlsxValue pairs to decode AI responses sensitive
.rules.xlsxsales.rules.xlsxRule configuration only — no data values, reusablesafe to share
.decoded.*sales.decoded.xlsxAI output with original values restored sensitive
REFERENCE

Keys vs presets

The decoding key contains real values — keep it private. The rules preset describes only how to transform data; it's safe to share or commit to a repo.

REFERENCE

Tips & best practice

  • Anonymize every column that could identify a person, company, or project.
  • Use meaningful prefixes (CUSTOMER, PRODUCT, REGION) so the AI preserves context.
  • Never share .key.xlsx alongside anonymized data.
  • Re-use a rules preset across reports so the same value gets the same token over time.