RPA-Everything lets an AI agent explore a repeated browser, desktop, Android, or office workflow once, then turns it into a reviewed deterministic Python Skill. Run locally, inspect evidence, and schedule only after a supervised run.

I built RPA-Everything after seeing the same computer-use pattern break down in repeated work. An agent is useful for exploring an unfamiliar UI, but asking it to rediscover every click on every run is costly and fragile. The workflow here is deliberately split in two: use an Agent once to plan and explore, save a trace, then turn that trace into a transparent Python Skill. The Skill can be reviewed, run locally, and scheduled without another model call. Today it supports browser, desktop, Android, iPhone-assist, and office-file paths. Browser flows prefer DOM selectors; Android prefers UI nodes; desktop coordinate steps are flagged for review rather than treated as production-ready. I would value feedback on the solidification workflow, safety boundaries for external actions, and the cases where this model is genuinely more useful than an always-on browser agent.

One thing I'd love to see is a "step diff" view that shows what changed in the generated Python skill between runs, so I can quickly spot when an app update silently shifts selectors or UI flows. That would make reviews way faster and catch breakage before scheduling.

RPA-Everything

Automate anything once with AI — replay it without AI

Automate anything once with AI — replay it without AI