Get started

See it work

Try Omna on fake data in 30 seconds, and see exactly what the AI would receive. Everything below is real output from Omna 0.6.25.

2 min readTry it on fake data in 30 seconds

Step 1 — Hide a sample message

After installing, run:

sh
omna mask "Email jane.doe@example.com, call 212-555-0199, AWS key AKIAIOSFODNN7EXAMPLE"

What the AI would receive:

code
Email [EMAIL_1], call [PHONE_1], AWS key [SECRET_AWS_KEY_1]

The email, the phone number and the key are gone. In their place are masked labels the AI can still talk about.


Step 2 — Try the realistic style

Some people prefer the AI to see ordinary-looking data instead of labels. Add --realistic:

sh
omna mask --realistic "Email jane.doe@example.com, call 212-555-0199, AWS key AKIAIOSFODNN7EXAMPLE"
code
Email alice.bennett@example.com, call 312-555-0112, AWS key [SECRET_AWS_KEY_1]

The email and phone became believable fakes that can't belong to anyone (reserved example addresses and 555 numbers). Your fakes will be different names — but the same real value always gets the same fake on your computer. The key stayed a label on purpose — a secret never gets a fake that looks real. See Masking controls for when each style is used.


Step 3 — Use your AI as usual

Open Claude Code, ChatGPT or any covered tool and work normally. Paste a config file with a key in it, ask a question about a customer email — whatever you'd usually do.

In our own test, we asked Claude Code to read a file holding a fake AWS key and spell the key backwards. It spelled the masked label backwards — ]1_YEK_SWA_TERCES[ — proof the AI only ever saw [SECRET_AWS_KEY_1], never the key. The file on disk kept the real key.


Step 4 — Check what was caught

sh
omna log

One line per request: when, where it went, how long hiding took, and what was hidden — counted, never shown. Add --verify to check that no receipt was changed or deleted.

sh
omna report

A weekly summary from those receipts — the same sections as the sample report, for this one computer.


What's next