StateAct
A long-horizon computer-use agent research project that helps agent builders explore a program-state-first approach for producing and evaluating desktop/OS task automation results.
Tool overview
Based on the available evidence, StateAct is best judged as a promising research direction rather than a broadly validated, production-ready developer tool. The current sources are almost entirely X posts sharing the paper and the authors’ claims. That is enough to support that it is getting attention, that its core idea is “program state before pixels,” and that it reports a new OSWorld 2.0 SOTA, but not enough to show easy adoption, reliable reproduction, or proven operational savings for normal teams.
In practice, its value is as a design idea for people working on computer-use agents, GUI agents, and desktop automation research: rely less on screenshots alone and more on accessible program or system state for long-horizon decision making. It is not a general-purpose RPA suite, and not an out-of-the-box browser automation framework. A more accurate analogy is a research method and evaluation stance for GUI/OS agents, closer to a paper prototype than to productized tools like Zapier, UiPath, or Playwright.
On cost and adoption barriers, the evidence does not include official pricing, API fees, hosted service details, or commercial support.