herdr

Quick Take

herdr is a terminal-based agent multiplexer: one terminal that runs, watches, and coordinates several AI coding agents at once. Created by developer Oğuz Çelik and released as open source, it is written in Rust and ships as a single binary with no GUI, no account, and no telemetry.

The simplest way to describe it: if you have used tmux, herdr is that, rebuilt for the age of agents. tmux gives you panes and persistence but has no idea which pane is working or stuck. herdr adds that awareness.

What herdr Is Best For

• Running multiple agents at once: one control room for parallel AI work
• Monitoring agent state: a live view of which agents are blocked, working, done, or idle
• Keeping sessions alive: persistent sessions you can reattach to later
• Terminal-native workflows: real terminal panes, so full-screen TUIs render correctly
• Coordinating coding agents: works alongside Claude Code, Codex, and other terminal agents

How herdr Works

herdr sits around your agents rather than replacing them:

• Agent-state sidebar. A color-coded view shows each agent as blocked, working, done, or idle, so you can supervise several at a glance instead of tabbing between windows.
• Automatic detection. It recognizes many terminal agents by process name and output, with 15+ supported by default (Claude Code, Codex, GitHub Copilot CLI, Devin CLI, and more) and no per-agent configuration.
• Real terminals and panes. Each agent gets a genuine terminal you can split into panes, tabs, and workspaces.
• Session persistence. Sessions survive a closed laptop; reattach from another terminal later.
• A local socket API. Agents can create workspaces, split or zoom panes, read output, and subscribe to state changes through a local Unix socket.

Getting Better Results From herdr

Start small. Run two agents before you run five. The value is supervision, and supervision only helps if you can actually follow what each agent is doing.

Keep a human in the loop. A multiplexer makes it easy to launch a lot of parallel work. That is exactly when it is easy to lose track, so review what each agent does before trusting its output.

Use it for genuinely parallel tasks. herdr shines when tasks are independent, one agent writing tests, another reading docs, another reviewing a change. For a single linear task, one agent is simpler.

Honest Limitations

• It is a developer tool. herdr assumes terminal comfort and existing agent setups. It is not a beginner starting point.
• It coordinates, it does not think. herdr does not do the coding; the agents inside it do. Capabilities depend on the agents and models you run.
• Review is still on you. Running several agents in parallel multiplies output and mistakes. Approval gates and review matter more, not less.
• Young project. It is new and moving quickly, so check the repository for current supported agents, install steps, and license.

Alternatives and Related Tools

• Claude Code and Codex: the terminal coding agents herdr is designed to coordinate
• Gemini CLI: another terminal-based agent you can run alongside others

herdr is best understood through the agent multiplexer concept. See also herdr vs Claude Code, which explains why they are complements rather than replacements, and the AI coding agent glossary entry.