Site Agent Runtime · Interactive demo
See how an agent prevents mistakes
Choose a tab, press buttons 1, 2, and 3, and see what happens. Real Python code runs in the browser. No installation required.
Can two tasks run at the same time?
The system must block a conflicting task while allowing the first one to finish. Try it yourself.
Start with the first button.
Click “Start the task”.
Can an agent make a risky change without approval?
First, validate a regular plan. Then increase the risk and try to bypass required approval.
No real websites are modified.
Click “Validate a regular plan”.
Can stale data overwrite a newer change?
Two simulated agents read the same text. The first changes it. See what happens when the second tries to save an outdated version.
A test store compares version fingerprints, while State Guard records the conflict.
Click “Both agents read the text”.
Public GitHub repository ↗ · Automated tests ↗
These are isolated demos with no access to live websites. Published Python modules run in the browser with test dependencies. File locks across independent processes and other low-level mechanisms are verified separately by GitHub Actions tests.