A layered check that escalates from fast and certain to rich and contextual.
Four checks run on every message, on the device, before it is sent. Together they catch what a simple keyword match would miss.
Known-shape secrets
Credentials, API keys, card numbers and anything with a recognizable, near-certain form. These are the cases that warrant a hard block rather than a warning.
Company-specific terms
Project codenames, customer names, unreleased product names. The organization seeds these in advance, and Alesia matches them on the device.
Resemblance to confidential docs
Catches paraphrased or reworded versions of material the company marked confidential. A plain keyword match would miss this entirely.
A judgment layer for the rest
Genuinely ambiguous cases get a contextual pass. Several small hints together can matter more than any single signal alone.
Four layers, one verdict
Each layer runs locally and contributes to a single decision: does this message look risky? Layers are independent, so a clear pass on one never overrides a strong signal from another.
Your terms, your documents
Detection quality depends on what your organization provides. Seed term lists and confidential documents through the admin console; the device matches against them at send time.
- Credentials matched against seed lists
- Document fingerprints stay on the device
- No content leaves to train or tune
Fast enough to feel invisible
Detection runs in milliseconds on normal hardware. The employee notices nothing until something genuinely looks risky.
Detection, in detail
No. All four layers run locally. Only the category, severity, and the employee's decision are shared with your organization.
Explore the rest
Stop sensitive data walking out through AI tools.
Request access for your security team. We'll set up policy seeding and a deployment plan that respects the privacy boundary.