Settings
Tune the loop to match how you build.
Loop defaults
Quality target
Loop converges at or above this score.
Max iterations
Hard cap on self-improvement attempts.
Autonomy
How much to defer to the agents.
Model
Reasoning model used by planner/reviewer.
Notifications
Loop converged
When a run hits the quality target.
Iteration failed
When an iteration can't make progress.
Reviewer confidence low
When reviewer confidence drops below 0.75.