Herdrm shows every coding agent on every machine in one Mac window

Herdrm, a native macOS console for the herdr coding-agent runtime, has reached 663 GitHub stars and 51 forks since developer missuo created the repository on 19 August. The counts come from the GitHub REST API, retrieved 7 September.

The app puts every workspace, coding agent and shell that herdr manages into one sidebar, covering the local Mac and any machine reachable over SSH. It lists Claude Code, Codex, Gemini, Grok and OpenCode among the agents it recognises. Selecting one attaches to the agent's live terminal with native text selection and full keyboard control, so the session looks and behaves the way it does in a plain terminal window. A system notification fires when an agent is waiting for input, and a search shortcut jumps to any agent on any connected device. Devices connect over SSH keys, Tailscale SSH, or a password stored in the macOS Keychain, and reconnect on their own after a drop. A two-pane file manager moves files between the Mac and any configured SSH device, and files or images pasted into an agent's pane arrive on the machine where the agent runs.

Built on a runtime with 35,893 stars

Herdr itself is a background server that owns each agent's terminal, tracks which one is working, blocked or done, and keeps sessions alive across disconnects. Its repository stands at 35,893 stars, read from the GitHub API the same day, and the project's site says it detects 21 agent command-line tools. Herdrm is one of several community clients forming around it: a code-review sidebar for herdr is at 613 stars, a read-only file viewer at 533, a terminal annotation tool at 382, and the herdrm README credits an iOS client as well.

macOS 14, early stage, and one default to check

Herdrm requires macOS 14 or newer, installs through Homebrew or a downloadable app bundle, and updates itself after that. The README labels the project early stage without full test coverage, and every machine it watches has to be running herdr. One default is worth knowing before a first run: when herdrm starts a new agent, the agent's permission-bypass flag is on by default, so the agent executes without confirmation prompts until that setting is changed.

The problem it addresses shows up as soon as agents run for hours at a time. A developer with Claude Code on a laptop, Codex on a desktop and a third agent on a rented server supervises all three by cycling through terminal tabs and SSH sessions, and learns that an agent has been waiting on a question only by going to look. One window that shows which agent needs attention, and sends an alert when one does, turns that checking into something the machine does. The permission-bypass default deserves the same attention: an agent that never asks is also an agent nobody stops before it acts, so a team should choose that flag on purpose. 663 stars in 18 days for a client this specific, sitting on a runtime near 36,000, says the supervision problem has a real audience.

Source

Primary source: missuo/herdrm on GitHub. Star and fork counts from the GitHub REST API, retrieved 2026-09-07. Runtime details from herdr.dev.