The orchestrator calls run_workflow with an inline spec: a read-only fanout of three reviewers, then one judge that merges their findings.
The ultra overlay shows every agent as a row with status, model, live action summaries, token usage, and elapsed time.
Reviewers finish at their own pace; the judge starts when all are done; the tool result carries the schema-checked verdict.
The orchestrator quotes the verdict in one sentence.
Protagonist: ultra.
Supporting: footer with the running cost.
Point: parallel sub-agents with structure and visibility, returning one checked result from one tool call.
# ultra demo
1. You ask for a review of three files.
2. The orchestrator calls `run_workflow` with an inline spec: a read-only fanout of three reviewers, then one judge that merges their findings.
3. The ultra overlay shows every agent as a row with status, model, live action summaries, token usage, and elapsed time.
4. Reviewers finish at their own pace; the judge starts when all are done; the tool result carries the schema-checked verdict.
5. The orchestrator quotes the verdict in one sentence.
Protagonist: ultra.
Supporting: footer with the running cost.
Point: parallel sub-agents with structure and visibility, returning one checked result from one tool call.