Running one AI coding agent is manageable. Running three, five, or ten simultaneously across different projects and machines is a coordination problem that traditional terminal tools were never designed to solve. Herdr is an open-source, Rust-built terminal multiplexer that treats AI coding agents as first-class citizens, giving developers a single window to manage, monitor, and orchestrate multiple agent sessions with persistent state.
Created in March 2026, Herdr has already gathered over 36,000 GitHub stars and more than 767,000 installs. It runs as a single binary on macOS, Linux, and Windows, licensed under Apache 2.0. The project raised a $6 million seed round and is actively hiring, signaling that the team plans to build around this tool long-term.
The Problem Herdr Solves
AI coding agents like Claude Code, Codex, Cursor, and OpenCode each run inside a terminal session. When developers work with multiple agents at once, a few pain points emerge quickly:
- Sessions die when the terminal closes, the laptop lid shuts, or an SSH connection drops.
- There is no way to see which agent is actively working, which is blocked waiting for input, and which has finished, without switching between terminal tabs or panes one by one.
- Agents cannot communicate with each other or coordinate work across panes.
- Running agents on remote machines requires separate SSH sessions with no unified view.
Traditional multiplexers like tmux solve the persistence problem for generic terminal sessions, but they have no concept of what is running inside a pane. Herdr builds on the tmux model and adds agent awareness, a structured sidebar, and a programmatic API that agents themselves can use.
Core Features
Persistent Background Sessions
Herdr runs as a background server process. The terminals live inside it, not inside a visible application window. When a developer closes their laptop, drops a network connection, or shuts the terminal client, the agents keep running on the server. Reattaching is instant with herdr attach. After a machine restart, Herdr restores the saved layout and can resume supported agent sessions from their last state.
Real-Time Agent State Tracking
This is the feature that distinguishes Herdr from tmux or Zellij. Herdr reads every terminal pane and detects AI agents running inside them using process-name matching and terminal output heuristics. Each detected agent is displayed in a sidebar with a live status indicator:
- Working: the agent is actively reasoning, generating code, or running commands.
- Blocked: the agent has stopped and is waiting for user input, such as a permission prompt or a question.
- Idle: an open shell ready to accept a prompt.
- Done: the agent completed its task and exited cleanly.
When an agent becomes blocked, Herdr surfaces that immediately so the developer does not have to check each pane manually. A glance at the sidebar tells the whole story.
Multi-Machine Management
Herdr can connect to remote machines over SSH and display their workspaces and agents alongside local ones in a single window. Each machine reconnects independently, so dropping one SSH connection does not affect the others. Developers can switch between machines without opening separate clients.
This is particularly useful for teams that run heavy workloads on cloud servers or GPU instances while coding locally. The agents on the remote machine keep working even when the developer disconnects.
Agent-Native API
Herdr exposes both a CLI and a local Unix socket API that agents themselves can use programmatically. Through this API, agents can:
- Spawn new panes and start other agents.
- Prompt other agents and wait until they are genuinely blocked.
- Read output from other agent panes.
- Subscribe to state change events.
This turns Herdr from a passive container into an orchestration layer. Instead of an agent firing keystrokes blindly into another pane, it can send a structured prompt, wait for a real response, and read the result. This matters for multi-agent workflows where one agent reviews code while another writes tests.
Plugin System and Marketplace
Herdr supports local executable workflow plugins with manifest actions and event hooks. The community has already published over 1,000 plugins. Developers can author plugins to extend pane behavior, integrate with external tools, or automate repetitive workflows. A marketplace for discovering and sharing plugins is available through the Herdr website.
Mouse and Keyboard, Both First-Class
Unlike tmux, which is keyboard-centric, Herdr is mouse-first. Developers can click panes, drag borders, split windows, and switch sessions from right-click menus. Keyboard shortcuts are available too, with tmux-style prefix keys (Ctrl+B). The tool adapts to whichever interaction mode the developer prefers at any given moment.
Herdr vs. tmux: Key Differences
For developers already comfortable with tmux, the transition to Herdr is straightforward. The pane model, detach/reattach workflow, and prefix key system are familiar. But several important differences set them apart:
- Agent awareness: tmux has no concept of what runs inside a pane. Herdr detects AI agents and tracks their lifecycle state in real time.
- Mouse support: tmux is keyboard-driven by default. Herdr treats mouse interaction as a primary input method.
- Socket API for agents: Herdr’s API lets agents control the multiplexer programmatically. tmux’s scripting interface is designed for human-operated automation, not agent-to-agent communication.
- Multi-machine view: Herdr unifies local and remote machines in one client. tmux requires separate sessions per machine.
- Session restore after restart: Herdr can restore layouts and resume agent sessions after a machine restart. tmux session state is lost when the server process dies.
tmux remains a solid choice for general terminal multiplexing. Herdr specifically targets the developer workflow around AI coding agents, where agent state visibility and inter-agent coordination matter.
Supported Agents
Herdr detects 21 AI coding agents out of the box, including:
- Claude Code
- OpenAI Codex
- Cursor Agent
- OpenCode
- Grok CLI
- GitHub Copilot CLI
- Gemini CLI
- Hermes
- Pi
Detection for most of these agents is zero-configuration: launch the agent inside a Herdr pane and it appears in the sidebar automatically. Herdr does not wrap or replace these tools. It simply owns their terminal sessions, adding persistence and state visibility on top.
This also makes Herdr a practical companion for developers experimenting with newer agents like Meta Muse Code or self-hosted options such as OpenClaw, as long as the agent runs in a terminal.
Installing Herdr
Herdr is a single Rust binary with no runtime dependencies. Installation is one command on each platform:
macOS and Linux:
curl -fsSL https://herdr.dev/install.sh | sh
Homebrew (macOS):
brew install herdr
Windows (PowerShell):
powershell -ExecutionPolicy Bypass -c “irm https://herdr.dev/install.ps1 | iex”
After installation, starting Herdr is a single command: herdr. The tool launches its background server, opens the TUI, and is ready to accept panes and agents. Developers coming from tmux will find the default keybindings familiar, while newcomers can rely entirely on the mouse-driven interface. Full installation options, including Nix and manual binary downloads, are available in the official documentation.
Quick Start Workflow
A typical first session with Herdr follows this pattern:
- Run herdr to start the server and open the interface.
- Split panes using the right-click menu or keyboard shortcuts.
- Launch an AI coding agent (e.g., claude for Claude Code) in one of the panes.
- The agent appears in the sidebar with its current state.
- Launch additional agents in other panes for parallel work.
- Press Ctrl+B then Q to detach. The agents keep running.
- Run herdr again later to reattach and check on progress.
For remote machines, developers add a machine with herdr machine add <hostname> over SSH. The remote workspaces and agents then appear in the same sidebar as local ones. The official quick start guide walks through each step in detail.
The Ecosystem Behind the Growth
Herdr’s rapid adoption, from zero to 36,000 stars in roughly six months, reflects a real gap in the developer tooling landscape. As AI coding agents move from occasional experiments to daily workflow tools, the infrastructure around them needs to evolve. Developers are not running one agent on one project anymore. They are running multiple agents across multiple repositories, sometimes on multiple machines, and they need visibility into all of it.
The project’s Y Combinator backing and $6 million seed round suggest that the team sees Herdr as more than a side project. The combination of an open-source core, an active plugin ecosystem, and a planned cloud offering (Herdr Cloud, currently on a waitlist) points to a platform strategy built around the agent runtime concept.
Frequently Asked Questions
Is Herdr free to use?
Yes. Herdr is open-source under the Apache 2.0 license. There is no paid tier, no account requirement, and no telemetry. The source code is available on GitHub.
Does Herdr replace tmux?
Herdr can replace tmux for developers who primarily work with AI coding agents. It uses the same pane and session model but adds agent state detection, mouse support, and a socket API. For general-purpose terminal multiplexing without agents, tmux remains a capable tool.
Can Herdr run on a remote server?
Yes. Herdr’s background server mode keeps sessions alive on any machine. Developers can attach remotely over SSH from their local terminal, a laptop, or even a mobile device. The agents continue running on the server regardless of whether anyone is attached.
What happens to agent sessions after a machine restart?
Herdr restores saved layouts after a restart and can resume supported agent sessions. The original terminal processes do not survive the restart, but Herdr replays pane history and brings agents back to their previous state where the agent CLI supports it.
How does Herdr detect agents?
Herdr uses process-name matching and terminal output heuristics to identify running AI agents. For most of the 21 supported agents, detection is automatic with zero configuration required.
Looking Ahead
The AI coding agent space is evolving fast. New agents launch regularly, and developers are finding ways to use multiple agents in parallel to cover different parts of a codebase. Tools like Herdr, alongside open-weight models like DeepSeek V4, represent the infrastructure layer that makes multi-agent workflows practical at scale.
For developers who have hit the limits of tmux or manual pane management when running AI agents, Herdr offers a purpose-built alternative that takes agent coordination seriously. The project is young, the ecosystem is growing, and the core tool is already functional for daily use.
