Agent Teams
Claude Code supports multi-agent teams where specialized agents collaborate on complex tasks. Cogpit gives you full visibility into these teams — see who's doing what, track task progress, and monitor inter-agent communication in real-time.
Due to Claude Code limitations, agent teams work best when you start the session from a terminal and use Cogpit as a monitoring dashboard. This gives you full visibility into all agents without interfering with their work.
Try it yourself
Explore the teams dashboard below — members grid with roles and active tasks, kanban task board with dependency tracking, and a real-time activity timeline.
Members
Tasks
Activity
Starting the implementation. I'll coordinate tasks — check TaskList for your assignments.
Auth system complete with JWT + refresh tokens. All tests passing.
Great work! Can you also add the settings page when you're done with the dashboard?
Found existing patterns in src/lib/hooks. The useQuery hook already handles caching.
Dashboard layout is done. Moving on to the interactive components now.
Went idle
Assigned task #5 to backend-dev: Add WebSocket support
How it works
Create a team
A team lead agent spawns the team using TeamCreate,
defining the project scope and creating a shared task list. Teammates are spawned with the
Task tool, each with a role and model assignment.
Assign tasks with dependencies
Tasks are created with TaskCreate and assigned to teammates.
Dependencies between tasks are tracked automatically — blocked tasks can't start until their blockers complete.
The kanban board shows this with
Blocked by #N
and
Blocks #N
badges.
Agents collaborate
Teammates work independently on their assigned tasks, sending messages via
SendMessage to coordinate.
The activity timeline shows all communication — plain chat messages, task assignments, idle notifications,
and shutdown requests — with color-coded senders and timestamps.
Monitor everything in Cogpit
Cogpit connects via SSE for live updates. Click any member to jump into their individual session. The team lead is marked with a crown icon. Active tasks show a spinning indicator with the task's current activity.
When to use teams
Large features
Split a full-stack feature across frontend, backend, and testing agents. Each works in parallel with task dependencies keeping things in order.
Research + implement
One agent researches APIs and patterns while another implements. The researcher feeds findings to the developer through messages.
Refactor at scale
Multiple agents each refactor different modules simultaneously. The team lead coordinates and reviews while specialists do the work.