Terminal dashboard manages four AI coding agents

A new terminal tool called agent-manager puts Claude Code, Codex, OpenCode, and Grok Build sessions into one hierarchical tree, organized by project, per its GitHub page.

Each agent still runs in its own tmux pane, so a session keeps working even after you close the manager. That matters if you're running more than one agent at a time. The tool tracks six states per session: working, waiting, finished, errored, idle, and dead. A stalled agent shows up instead of sitting silent until you happen to check.

The interface adds a spacebar shortcut for sending a quick prompt without attaching to a pane. It also has a full-screen diff view with syntax highlighting and line comments, for reviewing what an agent changed. It reports CPU, RAM, swap, disk, and network use per agent and across the whole fleet. That's useful once four agent CLIs are running on the same machine at once.

The more interesting design choice is buried lower in the feature list: agent-manager exposes its own session-management commands to agents over MCP. The dashboard view is the easy part. Giving agents MCP access to the session manager is the part that could let one agent spin up or kill another's session, unwatched.

It's a small project: MIT-licensed, 29 stars, built by a single GitHub contributor, YoanWai. Whether it holds up under real multi-agent workloads, the repo doesn't say.