Skip to content

AIgency scheduler

The scheduler repeatedly finds structurally valid, dependency-satisfied tasks and starts them within concurrency and capacity bounds.

AgentDock Pro

AIgency is available only in AgentDock Pro.

aigency-service.ts implements the scheduler as a repeated pass rather than a persistent loop: whenever a task settles (completes, fails, or otherwise changes state), the very next scheduling pass re-evaluates what's now eligible to run.

The gates a task must clear

  1. Dependency validation — every prerequisite must be genuinely completed, not merely no-longer-blocking.
  2. Pause gate — nothing new starts while the run is paused.
  3. Concurrency gate — both the run's global concurrency ceiling and any per-agent ceiling must have room.
  4. Capacity gate — a task whose target agent is mid-capacity-wait isn't started.

Verified bounds

BoundValue
Default concurrency2
Maximum configured concurrency6
Maximum retries per task5
Maximum fallback agents per task2
Interaction expiry24 hours
Mid-run silence stall warning15 minutes

These are hard caps enforced in the scheduler itself, not just UI-suggested defaults — configuring a higher value than the cap clamps to it rather than erroring.