Analyze restricted spreadsheets
Forecast and compare. The real figures never leave.
External AI on data you can’t share
The model sees encoded data. You get the real answer.
Let external AI work on data that can’t leave your environment. The analysis still holds, even when the answer depends on the values you protect.
- Design partners
- Finance and healthcare
- Deployment
- On-prem or in your cloud. Use our interface or the API
- Answer quality
- Same or better answers than the model’s own interface
- Works with
-
- and more
Every workaround gives something up: the data, or the answer.
The analysis that got waved off. The spreadsheet nobody would paste. The “we can’t use AI for that.”
| Approach | What the model receives | What you get back | The catch |
|---|---|---|---|
| Nothing | Nothing | The work doesn’t happen. | |
[CLIENT_1] [AMOUNT_1] |
Answers that hold up for names and IDs | Breaks when the answer depends on the values. | |
| Your raw data, covered by a promise | A full answer | A promise, not a control. It can’t stop a leak. | |
| Your raw data, on your own servers | Answers from a model you can host | Costly to run, and no access to the latest models or agents. | |
 |
Encoded data it can still reason on | The full answer from the latest models, decoded in your environment | You give up nothing. |
Example
Only encoded data leaves. The useful answer comes back.
Example · Aster Labs asks which market to prioritize
A useful answer is restored with its original context.
Click any card to see it in fullTap any box to see it in full
The model only ever worked on the right-hand side.
Not just calculations. A full analysis back, in your real figures.
The workbook is Encoded before it’s sent. The model works on the encoded copy.
A multi-page analysis, decoded in your environment. The highlighted figures are your real values, restored after the model replied.
Use cases
Three workflows you no longer have to give up.
Built for teams using external AI on regulated data.
Work with records containing PII or PHI
Summarize and classify case files. Identifiers stay encoded.
Give AI agents secured access
Agents work across your systems without exposing raw values.
1 / 3
Security and quality
We would rather be tested than believed.
The parity test: tens of thousands of checks against the same model used through its own interface The protected answer scores the same or better. How we test
- The model gets nothing to work back from. Before anything leaves, each protected value is transformed by a mathematical algorithm with randomization. The decode key stays only in your environment. The model receives neither your figures nor a lookup back to them. Your provider’s API logs show exactly what arrived: compare it with your original file.
- Everything that touches your real data, on-prem or in your cloud. Detection (ours, your own rules, or your existing detection system), encoding, decoding and the decode key. You can also add any entity by hand. Encoded prompts pass through flaxmor’s routing service to your model provider, using your own API key, so flaxmor never sees your real data.
- Negligible. Latency and token overhead are negligible compared with a direct API call, and token use is lower on large Excel workflows.
Design partnership
Make flaxmor the way your whole company uses external AI.
A long-term partnership, built around your workflows.
-
Start where AI is blocked today
The workflows where sensitive data is essential to the result.
-
Open it to every employee
Move your current AI calls through flaxmor: the same models, with sensitive values encoded before anything leaves.
-
Check it at every step
Read the provider’s logs and review the restored answers with your team.
-
Shape what we build
Your feedback shapes what we build next.
You’ll work directly with the people who built it.
-
Royee Flax
CEO & co-founder
Elite intelligence unit veteran with experience across multiple intelligence-gathering methodologies.
-
Dvir Mor
CTO & co-founder
AI engineer with cyber and quantum expertise developed through military intelligence service.
Let the answer take flight.
Skeptical? Good. Bring one blocked workflow and we'll show you exactly what the model receives.