What it does
Real-time mode can execute actions labelled read below the ordinary confidence threshold. Two finish choices can produce a completed status without a positive goal check, so completion needs a separate check.
Agent tooling
System One Harness
Drive an environment from a finite action menu, with Jev choosing the next step and code executing it.
Bookmark: System One Harness
Keep this in your collection.
Leave a noteWhat would you try with this?
: System One Harness
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
A developer can start with the pretend order workflow and define a small menu of actions for another task. Tell it what still needs to happen to reach the goal. Check the final result yourself, and label each action’s risk accurately: those labels affect when the program will act.