K-codeAgent Hub & running agents
When an agent fans out into subagents (parallel tasks, workflow steps, delegated reviews), K-code tracks each one. Agent Hub is where you watch them.
Open it
- In a thread, open the right panel and choose Agents, or press the keybinding for Agent Hub: Toggle (assign one in Settings → Keybindings; it has no default).
- For a project-wide view open the Running agents page (
/agents) from the command palette (Open running agents) or theagents.openkeybinding (assign one in Settings → Keybindings). It lists every live subagent across the threads of the selected project and environment, grouped by parent thread.
The grid
Each running agent gets a pane:
- the agent name, model and status (queued, running, done, failed, stopped)
- the current tool and a progress strip while it works
- the latest thinking excerpt and its replies, rendered as markdown
- token usage once the provider reports it
- a Stop button when that agent can be stopped (see Stopping subagents)
Layout controls:
- The grid is responsive, one to four columns, with no cap on the number of panes.
- Live / All filters to agents that are still running, or shows finished ones too.
- Maximize a pane to read its full log; restore to return to the grid.
- Live agents sort first.
Batch tasks
Some providers launch several tasks in one tool call. K-code splits those into one entry per task (for example “agent 2 of 5”) with its own status, progress, usage, model and result, and keeps the entries stable while progress arrives. Oh My Pi batch runs are supported this way.
Transcript detail
Open a pane’s log for the agent’s transcript. With full transcript mode on, logs show full thinking and full tool input and output instead of excerpts.
Mobile
The mobile app has an Agents sheet on the thread screen with the same information: live-first ordering, a batch hint, and a live progress strip per agent.