How to manage multiple coding agents running in parallel
Run each agent in its own Herdr pane, connect each machine with PeerBand, and use the dashboard or your MCP client as one live view of all of them. Status changes stream in as they happen, and webhooks tell you when one needs you.
The problem with many agents
One agent is easy to watch. Five agents in five terminals, across two machines, means constant tab switching to find the one stuck on a question. The work isn't running agents, it's noticing when one needs you.
One view of every agent
- Run your agents in Herdr. It detects each agent and tracks whether it is working, idle, blocked or done.
- Run
peerband setupon every machine with agents. - Open the dashboard. Every machine and agent appears with its live status, label and workspace.
The connector streams status changes to PeerBand instead of being polled, so the view updates the moment an agent stops or starts waiting, and your panes don't jump around.
Ask across all of them
From ChatGPT, Claude or Claude Code, ask "Which agents are blocked, on which machine?" or "Summarize what each agent finished in the last hour." The client calls list_machines, list_agents and get_agent_output across your account.
Only hear about the ones that need you
Add a webhook with the status filter set to blocked, and optionally idle, to get one Slack or Discord message per agent that needs attention. Agents can also send a note when they finish, with peerband notify.
Questions
Does PeerBand start or orchestrate agents?
No. You start agents in Herdr however you like. PeerBand gives you, your clients and your teammates a view of them and a way to message them.
Can I limit which agents clients see?
Yes. allowedAgents and allowedWorkspaces in the connector's connector.json restrict what MCP clients, the dashboard and events can see on that machine.