You're running three coding agents at once. One is building out the API layer. Another is writing tests. A third is fixing that CSS mess you've been ignoring for weeks. Which one needs you right now?
With tmux, you'd be squinting at six panes trying to remember which terminal belongs to which task — tabbing through windows hoping to catch a "waiting for input" prompt before the agent times out. It's manageable for one agent. For three, it's a mess.
herdr solves exactly this.
What It Is
herdr is a terminal multiplexer — think tmux — rebuilt from scratch with AI coding agents as the primary user. It's a single Rust binary around 10MB in size. No GUI, no Electron, no account required. Each agent gets a real terminal, not a wrapper's simulation of one, so even full-screen TUIs render correctly. And the sidebar tells you, at a glance, which agents are 🔴 blocked, 🟡 working, 🔵 done, or 🟢 idle.
That last part is the whole product.
Why It's Worth Your Time
Here's the thing about tmux: it was built before coding agents existed. It has no concept of what's running in a pane — it just renders characters. To get any agent state visibility with tmux, you'd wire up bells, custom hooks, and per-harness integrations. And even then, you'd have a patchwork that breaks whenever an agent changes its output format.
herdr does this with zero config. It uses process-name matching plus terminal-output heuristics to detect agent state automatically. Claude Code, Codex, Cursor, Hermes, Amp, OpenCode, Devin, GitHub Copilot CLI, and a dozen others are supported out of the box. If your agent isn't on the list, herdr still runs it — you just get multiplexing without the state detection.
The other thing that matters: nothing dies when you disconnect. A background server keeps every pane and agent alive. Detach from your laptop, walk away, reattach from another machine — or your phone over SSH — same session, same agents, still running. This is tmux's original killer feature, and herdr carries it forward intact.
herdr --remote ssh://you@yourserver turns your local terminal into a client of a remote herdr server. Clipboard paste into agents works over this connection — the thing plain SSH + tmux breaks because there's no way to pass the pasteboard across the wire.
Hands On
Install:
curl -fsSL https://herdr.dev/install.sh | sh
Also available via Homebrew (brew install herdr), mise (mise use -g herdr), or Nix (nix run github:ogulcancelik/herdr). Linux and macOS binaries are on the releases page.
Start it:
herdr
That's it. herdr checks for a running background server and starts one if needed, then opens a workspace. Drop into a pane and run your agent.
Keybindings follow the same prefix model as tmux — ctrl+b then the action key, so muscle memory transfers:
ctrl+bthenvor-— split the current panectrl+bthenc— new tabctrl+bthenShift+N— new workspacectrl+bthenw— switch workspacesctrl+bthenq— detach (agents keep running in the background)ctrl+bthen?— show all bindings
Mouse works throughout. Click to focus panes, drag borders to resize. For a terminal multiplexer, this is genuinely pleasant.
Connecting to a remote server:
herdr --remote workbox
herdr --remote ssh://you@yourserver:2222
If you're running agents on a VPS or a more powerful machine and want to supervise them from your laptop — or your phone — this is the flag you want. The local terminal becomes a client of the remote herdr server. It's not screen sharing; it's the actual terminal session.
Honest Verdict
It's v0.4.0 and it shows in places. Windows support is beta. The AGPL-3.0 license is a real consideration for organizations — commercial licenses exist, but it's a conversation you'd need to have with your legal team before using it in a company setting. And there's no revenue behind the project yet; it's one person building in the open and asking for sponsors. That's the longevity question every early adopter has to sit with.
But the core works cleanly. The agent-state sidebar is the feature that tmux will never ship because it would require tmux to know what you're running, and that's not in its design charter. herdr was designed from day one assuming the primary use case is humans supervising AI agents — and that constraint produces a sharper tool.
If you're running more than one coding agent regularly, the evaluation cost here is nearly zero: one small binary, one command, no accounts, no configuration. Try it for a week and see if you go back.
Go Try It
curl -fsSL https://herdr.dev/install.sh | sh
herdr
Open a few panes, start your agents, and watch the sidebar. You'll immediately understand what it's for.