herdrdev/herdr

The problem

Running several coding agents (Claude Code, Codex, Cursor, opencode) at once means juggling tmux windows or plain shells. When you close the laptop or lose SSH, the session dies with it. And with half a dozen agents working in parallel, the real cost is finding the one that has stopped and is waiting for your answer. herdr is a Rust-based runtime that owns the terminal panes your agents run in and labels every pane as working, blocked or idle.

What's different

Compared with tmux or Zellij, herdr does not try to be a general-purpose multiplexer with a scripting layer; it assumes every pane may hold an agent. That buys three things a plain multiplexer does not offer: a background server that persists sessions and restores layout after a restart, a combined agent list spanning local and saved SSH machines, and a socket/CLI API that lets agents spawn panes, prompt each other and wait until a peer is genuinely blocked. It also does not wrap or replace the agent — it owns the terminal, so anything that runs in a TTY works. The caveat is that agent state detection is per-agent integration, not magic; support breadth depends on how each agent is wired up.

Who it's for

Developers already running two or more coding agents in parallel, especially across a laptop plus one or more remote boxes, will get immediate value. Anyone doing agent orchestration on a single machine and happy with tmux plus a few shell scripts should probably not switch — the gain is marginal. Teams that need a GUI, audit trails, or centrally managed agent permissions should look elsewhere; this is a keyboard-and-terminal tool.

Cost of adoption

One Rust binary, no Electron, Apache-2.0, installable via a shell script, Homebrew, mise or release binaries, with a Windows build marked beta. The runtime cost is a habit change: you move your existing agent workflows into herdr panes and remember detach/reattach (ctrl+b q). Risks: session restoration brings back layout and resumable agent sessions, but the original processes do not survive a machine restart and not every agent supports resumption; the remote-machines story is newer than the local one; the README is mostly links to herdr.dev/docs, so details like plugin API stability and agent integration coverage have to be checked in the docs.

Verdict

A 211-day-old project with roughly 45.7k stars and only +147 and +144 on the two days it appeared on the trending board — large accumulation, modest current velocity. Interest looks real but the spike is not explosive, and the age means the plugin ecosystem and agent integrations are still young. If you already run parallel agents, it is worth an afternoon; if you run one agent at a time, wait.