Sentry
Authentication that keeps going after you sign in. Sentry learns how a person types, moves and scrolls, and locks the workstation when the behaviour stops matching.

A password proves who you are once. Sentry keeps checking, quietly, for the rest of the session.
It builds a behavioural profile from the way someone uses their computer and compares live activity against it every few seconds. There is nothing extra to remember and nothing to type.
An unlocked laptop or a stolen password gives full access, and a traditional login never notices.
Continuous checks only work if they stay out of the way. Sentry had to be sharp enough to catch someone else at the keyboard, forgiving enough not to lock out its owner on an off day, and light enough to run in the background all day.
- Behavioural profileTwelve measures, including keystroke timing, mouse dynamics, scroll patterns and error-correction rate, combined into one signature.
- Background captureA daemon collects input events and recomputes the behavioural metrics every 10 seconds.
- Per-user modelAn Isolation Forest ensemble trained for each person, with contamination tuned to the size of their training data.
- Graduated responseAn OK, Risk and Lock state machine that escalates step by step and locks the workstation on clear anomalies.
- Desktop appElectron and React, talking to the Python backend over stdio JSON.
- Live controlsReal-time metrics and a sensitivity setting for each input channel.
Capture
Input timing and movement are recorded in the background and turned into features every 10 seconds.
Compare
Each new window of behaviour is scored against that person’s own model.
Respond
Normal behaviour stays OK, unusual behaviour raises a Risk state, and a clear mismatch locks the workstation.


