The agent map and records
The Agents card beside the chat lists every agent by what it needs. The lanes view lays out the whole run, and a record shows one agent in full.
Where in StudioChat → Agents
Updated View as markdown
On this page
The agent map is the Agents card that sits beside the chat during an Orchestra run. It keeps every agent in view while the conversation scrolls. Click a row to open that agent's record; click the expand button to see the whole run as lanes.
Where the map shows
- Beside the chat when the window is wide enough and the side panel is closed.
- As a drawer otherwise. Open it with Agents in the header of the composer's plan panel. It slides in over the conversation from the right; click outside it or press Esc to close it.
The Agents card
The header shows Agents and the number of agents in the run. When the orchestrator is waiting on you, a "needs you" pill appears; click it to jump to the question in the chat. Below the header are three sections:
| Section | What is in it |
|---|---|
| Needs you | Agents whose question the orchestrator has taken to you. The row says "Orchestrator asked you in chat" and how long the question stays open. Go to question scrolls to it. |
| Running | Agents at work, agents asking the orchestrator, and agents queued for a free slot. |
| Finished | Agents that are done, failed or stopped, folded to one line. Click it to unfold. |
Each row shows the agent's task, its status (Running, Asking, Queued, Done, Failed or Stopped) and its clock. When the orchestrator sends an agent a follow-up, the follow-up hangs under the first run's row.
You answer a question in the chat, to the orchestrator. The orchestrator then answers its agent.
The lanes view
Show all agents, the expand button in the card's header, lays the run out over the conversation:
- The Orchestrator sits on the left.
- Each plan step is a lane, in plan order. Without a plan, each kind of agent gets a lane. Agents that no step claimed are under Unassigned.
- Each card shows the agent's provider, name, model and status, its task, what it is waiting on and how its run went. A queued card says "Waits for a free slot".
- A Follow-up sits to the right of the run it continues, in the order the work happened.
Click a card to open its record. Choose Chat to go back to the conversation.
An agent's record
The record opens as a window over Studio. Use the switcher to move to the next agent without closing it.
The left column says who the agent is and how its run stands:
| Field | Meaning |
|---|---|
| Step | The plan step it works on. |
| Started | When it started. |
| Running for, Ran for, Waiting for | How long it has been working, worked, or waited. |
| Access | The execution mode it ran with: Read only, Workspace write or Full access. |
| Context | How full its context window is. Click it for the agent's Context usage. |
| Tokens, Cost | What all its runs used. See Limits and usage. |
| Runs | The First run and each follow-up, each with its own status. |
| Targets | The models it ran on, including fallbacks. |
The right column leads with what the agent waits on, how it moved between models, or why it failed. Then comes the Result: the agent's summary and findings, with Changed files, Checks and Blockers when it reported them. Two tabs follow:
- Transcript: the agent's whole conversation, including the task the orchestrator gave it, its reasoning and its tool calls.
- Questions & updates: the questions it asked the orchestrator with their answers, and the updates the orchestrator sent it.
Note
A record is for reading. You cannot answer, message or stop an agent from it. Talk to the orchestrator in the chat instead.
In the transcript
The orchestrator's own work shows as compact rows in the chat, such as Agent Started, Agent Finished, Question Answered or Agent Timed Out.