What and how
Task, solution, stack and status.
- Task
A desktop AI agent for Windows: the user writes a task in plain words and the agent carries it out on the computer. A personal R&D project.
- What was done
The agent runs in a loop: observe, plan, act, verify, replan.
- Working: plain-text task input (prototype), actions in the browser and on the desktop in selected scenarios.
- In development: on-the-fly planning, finding interface elements by meaning, checking results, recovering from errors.
- Planned: confirmation of risky actions (architecture in progress) and voice control.
- Stack
- C#, .NET, WPF
- Windows UI Automation, accessibility APIs
- Browser automation
- LLM APIs
- Built with Codex
- Status
Prototype, paused.
- Services
- Links
The repository has the architecture and docs; the code is not published.