Guides
Desktop AI Best Practices
Getting the most reliable results from a companion with real desktop access.
Be specific about scope. "Clean up my downloads folder" is more reliable than "organize my computer" — desktop AI works best with a bounded, describable target.
Let confirmation gates do their job. Reviewing what's about to happen before confirming a destructive action isn't friction to route around; it's the mechanism that makes automation trustworthy enough to actually delegate to.
Name your projects and keep folder structures reasonably conventional. Project understanding in the Coding Canvas works better with a project it can recognize than with an unlabeled folder soup.
Report what didn't work. An honest "that failed because X" from PawOS is more useful than a vague success — treat those messages as real diagnostic information, not an apology to skip past.