Step 1
Trigger
Webhook, schedule, queue or human handoff. Anything that fires an event.
Step 2
Plan
An agent breaks the goal into typed sub-tasks with budgets and constraints.
Step 3
Route
Each sub-task is dispatched to the optimal model, tool or sub-agent.
Step 4
Execute
Tools run in sandboxes; results stream back with structured traces.
Step 5
Verify
Guardrails, evals and policies score every output before release.
Step 6
Deliver
Results are returned, cached, and observable end-to-end.
Live runs
Every run, replayable.
00:00.012trigger.received
00:00.103plan.generated · 4 sub-tasks
00:00.418tool.web_search → 12 results
00:01.220model.gpt-5o · 1,204 tokens
00:01.812verifier.pass · score 0.94
00:01.901deliver.ok · 1.89s total