ProductsPrivacyLibraryDocsPricingGitHubAdd to ChromeDownload for Mac →
Two problems · one local solution

Mask the data.
Kill the token tax.
Before AI ever sees it.

The semantic layer between your data and every AI. Omna masks PII and cuts token waste before anything is sent.

Download for MacAdd to Chromepip install omna
Works withClaudeChatGPTGeminiPerplexityVS CodeVS CodeCursorCursorGrokGrokor any AI you use
Watch demo
Free during early access — no account, no credit card
Original filefinancial_report.csv240 MB · 12,000 rows
↓
100% local · on-device
↓
AI-ready outputfinancial_report_clean.csv12 MB · 240 rows
PII removed~95% smaller
↓
Send to any AI — masked & lean
ClaudeChatGPTGeminiPerplexityDeepSeekGrokGrok
Two problems · one local solution
The token tax

Most of what you send an AI is noise — and you pay for every token of it.

up to 99%fewer tokens sent
  • Lower AI bills
  • Faster responses
  • Sharper context
  • Fewer rate limits
PII exposure

Sensitive data should never reach a model unmasked. With Omna, it never leaves your machine.

0 bytesof PII ever sent
  • Automatic detection
  • Context-aware masking
  • 100% local processing
  • Audit trail included
Three ways to use Omna

One Omna engine. Three products.

The Mac app, browser extension, and Python library are different ways into the same local engine — your data stays on your device in all three.

Mac app

A floating capsule on your Mac. Drop in a file and Omna slices it to just the rows you need — PII masked — ready for Claude, ChatGPT or Cursor.

customers.csvfile dropped → sliced
Token Tax killed−77% tokens261 of 12,000 rows sent · 32 PII masked
  • Capsule + menu-bar controls
  • Slices & masks fully on-device
  • One-click drop into Claude, ChatGPT & Cursor
Download for MacFree · Apple-notarized · macOS 13 and up
Chrome extension

Type or upload in any AI chat. Omna masks PII and slices big uploads on the fly — leak nothing, send fewer tokens.

You type or uploadEmail jane@acme.co, phone (415) 555-0132
Omna masks it before sending ↓
AI receivesEmail [EMAIL_1], phone [PHONE_1]
  • Masks PII live as you type
  • Slices large file uploads too
  • Works on ChatGPT, Claude & Gemini
Add to Chrome — freeChrome · Edge · Brave · Arc · Safari & Firefox coming soon
Python library

Open source, built on Polars. Mask PII, search semantically, and query data in plain English — right in your dataframe.

.mask_pii().search().ask()
import polars as pl
import omna

df = pl.read_parquet("tickets.parquet")

# mask every PII entity, with an audit trail
clean, audit = df.omna.mask_pii()

# semantic search — runs locally
df.omna.search("billing problem", on="issue", k=5)
View on GitHubpip install omna · MIT
Drop in
CSVParquetPDFExcelWordJPEGPNGScreenshots& more
Audio & video coming soon
What leaves your machine?

Nothing sensitive. Ever.

Local-first by design: nothing leaves your device. Every kind of personal data is detected and masked on-device — each value swapped for a safe placeholder before a single byte is sent.

IdentityNames, facesJane Doe→[IDENTITY]
ContactEmail, phone, addressjane@acme.co→[CONTACT]
Government IDsSSN, passport, licence123-45-6789→[GOVT_ID]
FinancialsCards, accounts, IBAN4111 •••• 4242→[FINANCIAL]
HealthcareDiagnoses, prescriptions, insuranceType 2 diabetes→[HEALTH]
Digital footprintIP, cookies, device IDs192.168.1.24→[DEVICE]
EmploymentSalary, reviews, HR records$148,000/yr→[EMPLOYMENT]
Corporate secretsInternal codes, pricing, strategyProject Atlas→[INTERNAL]
CredentialsPasswords, API keys, tokenssk-live-9f2a…→[SECRET]
Custom patternsEnterprise-defined, per clientACME-04417→[CUSTOM]
Up to 99% smaller files & fewer tokens0 bytes sent to OmnaAudit-ready by default
$ pip install omna

A local-first Rust kernel slices and masks your data on-device — nothing ever leaves your machine. The Python library is open source (MIT).

View on GitHub
0bytes sent to Omna
100%local processing
FAQ

Is Omna really free?

Yes — free during early access. No account, no credit card, no trial timer.

Mac app or Chrome extension — which do I need?

The Mac app handles files and folders on your machine; the extension masks and slices right inside web AI chats. Most people use both, and the Python library for dataframes and pipelines.

What data actually leaves my device?

None of your raw data. Detection, slicing, and masking run locally in the Rust kernel — only the masked, slimmed text you choose to send reaches the AI. The apps send a small anonymous usage ping — an event name and a random device ID, never file contents, prompts, or personal data. Details in our Privacy Policy.

Which AI apps work today?

Claude, ChatGPT, Gemini, Perplexity and DeepSeek — plus Cursor and any other web app or API you paste into.

Is it open source?

The Python library is open source under the MIT license. The Rust kernel and Mac app are proprietary and free during early access.

Does it run on Windows or Linux?

The Mac app needs macOS 13 or later. The Python library runs anywhere Python and Polars do — Windows, Linux and macOS. Native Windows and Linux apps are in the works.