Agent Manager puts many coding AI agents in one keyboard-driven terminal list

What Agent Manager is

Agent Manager is a local terminal app for people who run several AI coding agents at the same time. It was posted to Hacker News as a Show HN, the forum's label for projects their makers present. According to the project page on github.com, it is meant to be a thin layer over command-line tools you already have, not a replacement for them.

The list of supported tools is long: Claude Code, Codex, OpenCode, Grok, Gemini CLI, Antigravity CLI, Pi, Command Code, Hermes Agent, Muse Code, Oh My Pi, Cursor CLI and Kiro CLI. They run side by side. The project says a tool it does not list yet counts as a feature request.

One list instead of a pile of tabs

The core idea is to replace hunting through terminal tabs. Every agent session shows up in a single list with a live status, arranged in a project tree you can fold and reorder. You can see at a glance which agent has finished and which is stuck.

You can also act without opening a session. Pressing space sends a prompt straight into a session, or starts a new agent in the selected group. A session that has died can be revived on its own conversation, and another key forks a conversation into a separate named copy. The mouse works too, including drag-and-drop to reorder rows. The project says a few tools, among them Hermes, Antigravity, Oh My Pi, Cursor CLI and Kiro CLI, cannot be forked.

A shell can be opened under any agent for builds, Git or one-off commands, so those tools sit in the same list.

How it keeps your setup intact

Each agent runs inside tmux, a long-standing terminal tool that keeps sessions alive in the background. That means sessions survive if you close the manager. The project says they live on a private tmux server named agentmgr, kept apart from any tmux you run yourself.

Each session launches your own installed tool unchanged. Your login, subscription, config files and MCP servers carry over. MCP, short for Model Context Protocol, is a standard way to give agents extra tools.

Sessions can also start in their own git worktree, which is a separate working copy of a repository on its own branch. That helps keep one agent's changes from colliding with another's. You can pick a model, reasoning effort and profile per session, and the project says those lists come from each tool at runtime.

Reviewing what the agents wrote

A key press opens a full-file, syntax-highlighted diff of what an agent changed. A diff is a side-by-side view of old and new code. You can leave comments on individual lines, and pressing a key sends all of them back to the agent as one review prompt.

The tool also lets agents work with each other. Any MCP-capable session can spawn another agent, message it and wait until it finishes. You can store named instructions as saved prompts and reuse them in several places.

One gap is stated plainly: cost tracking is not here yet. If you pay per use, you will still need another way to watch spending.

Installing it, and the caution that goes with it

The app runs on macOS and Linux, natively on Windows with a helper called psmux, and inside WSL2. On a Mac, brew install agent-manager is the short route. Other routes include an install script, and the project advises downloading the script and reading it before running it. The tool needs tmux 3.1 or newer and git.

The project is also looking for contributors and has a Discord community, a governance file and a private channel for security reports. Our own read: it was posted by its maker, has a tiny score on Hacker News, and has not been independently tested. Price is not stated, and we have not confirmed the licence.

There is a security angle too. Every extra agent is another place where a command can run. Try it in a throwaway repository first, keep permissions minimal, and read the code before pointing it at anything you value.