AbilityOS

A desktop mentor that helps people learn a job by doing it. It sits beside the learner, sees the task, and gives the smallest hint that helps without handing over the answer, even with no internet.

  • Rust
  • TypeScript

How it works

  1. Learner's windowwith consent
  2. Read the screenlabels and screen text
  3. Check the workrules first, model second
  4. Smallest hintnever reveals the answer
  5. Learner modelskills and forgetting
  6. Trainer viewwho is stuck now
With the learner's consent, the app reads the window they are working in. It checks the work against the task, rules first and a model only where rules cannot decide. It then picks the smallest hint that helps, updates a model of what the learner can do, and shows trainers who is stuck.

Decisions that matter

  • Rules judge first, a model only when needed

    Rule-based checks come first. A model is used only where rules cannot decide, and the system is allowed to decline to judge rather than guess.

  • Hints that cannot leak the answer

    Help comes in tiers, starting with the smallest nudge, and the ladder is built so that no hint reveals the protected answer. As the learner model shows skill growing, the help fades.

  • It works with no internet

    Coaching can run on a model on the learner's own computer, and one deployment mode is fully air-gapped. That matters for learners in correctional facilities, where it has to work offline.

  • It sees only the window being used

    The app keeps only what is in the learner's working window, never reads password fields, and organizations can limit what it watches.

Limits

Paused, with nothing deployed. Linux desktops are tested in virtual sessions; Windows has not run on real hardware yet, and macOS is not release-ready.

Next projectStatuta