herdr: Terminal Agent Multiplexer for AI Coding
ogulcancelik/herdrherdr is a terminal agent multiplexer written in Rust that keeps every AI coding agent you're running visible in one place. Think tmux, but built around the fact that you're now juggling several coding agents at once — each pane shows the real terminal, with the agent's state read as blocked, working, or done.
Reach for herdr if you're running several coding agents (Claude Code, Codex, and the like) at the same time and keep losing track of which one is stuck waiting on you versus still working. The detach-and-reattach-over-SSH model and the socket API that lets agents drive their own panes are the real draw. Skip it if you only ever run a single agent in one terminal, or if your organization can't take on an AGPL-licensed tool and its Windows build is still marked beta.
The problem it solves
Running one AI coding agent in a terminal is easy. Running four or five is where it falls apart: you tab between windows, lose track of which agent is blocked on a prompt versus grinding through a task, and if your SSH connection drops or the machine reboots, the sessions are gone. Plain tmux keeps the panes alive but has no idea what an agent is doing inside them, so you're still eyeballing raw scrollback to figure out state.
What is it?
herdr is an open-source agent multiplexer for the terminal, written in Rust and shipped as a single binary with no Electron. It runs your AI coding agents in panes you can split, detach, and reattach — including over SSH — and reads each agent's status as blocked, working, or done from the real terminal output rather than a wrapped interpretation. It also exposes a socket API so agents themselves can spawn panes, read output, and wait on one another. The repo is dual-licensed under AGPL-3.0-or-later with a separate commercial license.
Why it's getting attention
herdr has drawn about 15,374 stars, and the timing tracks with how many developers now run multiple coding agents in parallel. Its pitch lands on that shift: it's a terminal multiplexer that treats coding agents as first-class, not a generic tmux you bend into shape. The repo topics name Claude Code and Codex directly, it's one Rust binary instead of an Electron app, and the socket API that lets agents orchestrate other agents is the kind of primitive people building agent workflows go looking for.
Key features
- ✓Per-agent status read from the live terminal — blocked, working, or done — instead of a wrapped guess
- ✓Detach and walk away: agents keep running, and you can reattach from any terminal or over SSH, with sessions surviving restarts
- ✓A socket API that lets agents spawn their own panes, read output, and wait on each other (the repo links an 'agent skill' for this)
- ✓Both tmux-style prefix keys and mouse click/drag/split, usable interchangeably
- ✓A plugin marketplace for extending panes and workflows
- ✓Distributed as one Rust binary with no Electron, running inside whatever terminal you already use
Best use cases
- •Supervising several coding agents at once and seeing which are waiting on input versus still working
- •Kicking off long agent runs on a remote box, detaching, and reattaching later over SSH
- •Letting one agent programmatically spawn and coordinate other agents through the socket API
- •Replacing a hand-rolled tmux setup with a multiplexer that actually understands agent state
How to install / try
Install with the shell script: `curl -fsSL https://herdr.dev/install.sh | sh`. Package managers also work — `brew install herdr` or `mise use -g herdr` — and there's a beta Windows path via `powershell -ExecutionPolicy Bypass -c "irm https://herdr.dev/install.ps1 | iex"`. Prebuilt binaries are on the GitHub releases page.
How to use
Run `herdr` from the directory where the work lives, then split panes and start your agents in them. Press `ctrl+b q` to detach and walk away; run `herdr` again to reattach to the same session. The README points to a quick start and fuller docs at herdr.dev/docs for keyboard shortcuts, supported agents, configuration, and the socket API.
Strengths
- ✓Shows each agent's state at a glance rather than making you read raw scrollback
- ✓Sessions survive detach and restart, and reattach cleanly over SSH
- ✓A single Rust binary with no Electron, so it drops into your existing terminal
- ✓The socket API is a real building block for agents that coordinate other agents
- ✓Prefix keys and mouse both work, so you're not forced into one interaction model
Limitations & risks
- △It's AGPL-3.0-or-later; the project sells a separate commercial license precisely because AGPL is a blocker for some organizations, so check your compliance stance first
- △Windows support is still labeled beta in the README, so the smooth path is macOS and Linux
- △The README is thin on specifics — the supported-agent list, plugin details, and socket API docs all live off-repo at herdr.dev rather than being spelled out here
- △It's actively evolving: the repo frames sponsorship as funding 'stability' and 'the path to a real agent runtime,' so this is a maturing tool, not a finished runtime
- △If you don't already use a multiplexer, the tmux-style prefix keys carry a learning curve
Alternatives
Who should try it — and who should skip
Developers who routinely run several AI coding agents in parallel and want one terminal to see and steer them all — especially anyone starting long agent runs on a remote machine and reconnecting over SSH. If you only run a single agent at a time, or your organization can't adopt an AGPL tool with a still-beta Windows build, a plain tmux or Zellij setup is the safer pick.
Frequently asked questions
herdr is an open-source agent multiplexer for the terminal, written in Rust. It runs your AI coding agents in panes you can split, detach, and reattach over SSH, and it reads each agent's state — blocked, working, or done — from the real terminal output.
It's in the same family and borrows tmux-style prefix keys, but it's built around coding agents: it surfaces per-agent status and exposes a socket API so agents can spawn and coordinate their own panes, which plain tmux doesn't do.
Yes. You can detach from a session, leave agents running, and reattach from another terminal or over SSH, and the README says sessions survive restarts.
The repo topics reference Claude Code and Codex, and the README points to a supported-agents page in its docs for the full list, which isn't spelled out in the README itself.
The open-source build is licensed under AGPL-3.0-or-later. A separate commercial license is offered for organizations that can't comply with AGPL.