Herdr: a tmux-style terminal multiplexer built to wrangle coding agents

Herdr is a single-binary terminal multiplexer aimed squarely at people running multiple AI coding agents at once. Like tmux or Zellij, it offers split panes, tabs, and persistent sessions that survive the terminal closing — but unlike those tools, it understands agents, surfacing per-pane state (blocked, working, done, idle) at a glance across a whole session. Agents run inside real PTYs rather than a rebuilt chat view, and because sessions live on a server, you can start work on a Mac Mini or a sandbox VM, detach, and reattach later from another machine or even a phone over SSH.

The pitch is a middle ground between two existing categories: classic multiplexers that persist sessions but know nothing about agents, and desktop agent-manager GUIs that understand agents but are stuck on the machine running the app. Herdr keeps everything in the terminal, works with any terminal emulator (Ghostty, Kitty, iTerm, Alacritty), and deliberately ships with no Electron, no account, and no telemetry. A --remote mode installs itself on the host, bridges the local clipboard including image paste, and preserves keybindings. Stable builds target Linux and macOS, with Windows in preview beta.

What makes it more than a viewer is its control surface: a CLI and JSON socket API expose workspaces, panes, output, and blocking waits (e.g. herdr wait agent-status --status done), so agents can drive the multiplexer the same way a human does. That same API doubles as the plugin interface, with a community marketplace for notifiers, layout presets, and release checks. Herdr advertises out-of-the-box support for terminal agents including Claude Code, Codex, Cursor, Copilot, and Grok CLI, with richer state for integrated tools. It’s an early-stage independent project — the listed company logos represent individual engineers, not endorsements.