Privacy Layer · live demo
Use AI on customer data. Keep the personal details at home.
Hezico puts a privacy layer in front of the model. Names, IDs and contact details are swapped for placeholders before a prompt leaves your network, and swapped back in the answer. Try it below. Nothing you type leaves this page.
Your prompt
What the model sees
Answer you get back
The model's answer here is simulated in your browser so the demo works offline. In production, the masked prompt goes to the model you choose, such as Claude, GPT or a private model in your own cloud.
Masking log
Every swap is recorded, for your privacy records.
| Placeholder | Kind | Original (partly hidden) | Rule | Time |
|---|---|---|---|---|
| Run the layer to fill the log. | ||||
How it runs for you
Built into your environment.
Deployed in your own AWS, Azure or GCP account. The mapping between placeholders and real values never leaves it.
Put it in front of Claude, GPT or Gemini, or a private model hosted in your environment for the most sensitive work.
We add rules for your own identifiers, such as member numbers, policy numbers and student IDs, alongside names, NRICs, TFNs, Medicare numbers and contact details.
The masking log shows what was protected, when and by which rule, for privacy, PDPA and APP reviews.
What we tell every client
A privacy layer reduces risk. It doesn't remove it.
- Detection is never perfect. We test it on your real documents and tune it until misses are rare, and we report the miss rate.
- Context can still identify a person, for example a rare job title in a small town. For that work we recommend a private model.
- Masked data can still be personal information under the Privacy Act and PDPA, so your existing obligations stay in place.
Want to see it on your own data?
In 30 minutes we'll run it on a sample of your documents and show you what it catches, what it misses, and what a private model would change.