Orchestration Console

Live view of a multi-agent system: planner → specialist agents → human-in-the-loop

5 agents active

Agent Graph — "Quarterly Report" task

UserRequest Plannerdecomposes task Data AgentSQL + RAG lookup Analysis Agenttrend synthesis Writer Agentdrafts narrative Reviewhuman approval
planner: idle data_agent: running analysis_agent: queued awaiting_approval: 1

// trace_id: 8f2c-91ab

[planner] decomposed task into 3 subtasks

[data_agent] tool_call: query_warehouse(metric="revenue", range="Q3")

[data_agent] retrieved 4 rows · 18ms

[data_agent] tool_call: vector_search("Q3 retention notes")

[memory] context window updated · 1.2k tokens stored

[analysis_agent] waiting on data_agent output…

[guardrail] output contains financial claim → flagged for human review

[system] escalated to approval queue

Human-in-the-Loop Approvals
TaskFlagged byReasonAction
Quarterly report — financial claimGuardrail agentUnverified revenue figure

Privacy-safe recreation of a real build — visualizes multi-agent orchestration, memory, tool-calling and human-in-the-loop review for a production AI system. Interactive: the trace log streams live — approve or edit the flagged task to unblock the writer agent.