Every-30m Processes, Credit-Aware Scheduling, and Stronger Memory Retrieval

Processes can now run every 30 minutes, recurring work stops queuing when credits are depleted, and agent memory retrieval became more robust.

This update sharpens recurring execution. Schedules are more flexible, out-of-credit behavior is more explicit, and memory retrieval keeps getting better under real use.

A New every-30m Recurrence Pattern

Processes can now be scheduled every 30 minutes, which fills a useful gap between hourly automation and fully continuous work.

  • Higher-frequency recurring work is now first-class
  • Stale-run handling is tuned for the shorter interval so longer deep-work runs are less likely to be cut off too early

Recurring Work Now Stops Before It Wastes A Run

When an account is out of credits, the scheduler now skips spawning the recurring run instead of letting it fail later and more confusingly.

Result: credit exhaustion is easier to understand, and recurring workflows fail in a cleaner way.

Memory Retrieval Keeps Getting More Reliable

Agent memory retrieval and health handling were tightened again, which makes memory-backed planning chat feel more dependable when you return to older context.