Watch every running agent¶
The terminal client shows the whole agent tree while it runs, so you can follow every sub-agent without leaving the screen. Sub-agents covers how delegation itself works.
The fleet sidebar¶
When the root agent spawns sub-agents, a fleet section appears in the sidebar with one row per agent. Each row shows a status glyph, the agent's label, its model, its tool count, its elapsed time, and its token totals. A filled dot is running, a check mark is finished, and a cross is failed.
A row carries a tool count and never tool names, so it stays scannable. The panel lives in fleet_panel.py.
Rows tick on their own while the agents run. A stall shows as a readout on that agent's row, so a hung sub-agent stays visible even while the root keeps working.
Drilling into an agent¶
Select a row to open that agent's transcript in place, with its own breadcrumb and footer. Press Esc to return to the live root transcript. The fleet sidebar stays put. The drill in view lives in fleet_drill.py.
Orchestration cards¶
Fleet tool calls render as dedicated cards instead of raw JSON. The cards live in orchestration_cards.py.
- A spawn card names the task, the model, and the agent type.
- A fleet check renders as a small table listing each agent, its status, its tokens or steps, and its last tool.
- A tool discovery renders a short list of the tools it found, each with a one line description.
Everything streams in order¶
One hub subscribes to the session's event bus and demuxes every event by agent, so text, tools and spawns render in order. Sub-agent output streams live and labeled rather than arriving as a batch at the end. The hub lives in agent_transcript_hub.py.
Next steps¶
- The Interface covers the transcript, the plan dock, and approval prompts.
- Configuration covers flags, the config chain, and session recovery.
- Sub-agents covers how delegation works across every Mewbo surface.