Who Herdr is for#
Developers running long AI coding tasks overnight
Start an agent on a codebase refactor or test-generation task before leaving the office. Herdr keeps it running on a remote server or desktop. Check progress from another machine without reattaching until the agent is blocked and needs input.
Skip if:
Your tasks complete in under an hour and you never leave your desk during them. Session persistence only matters when task duration exceeds a single uninterrupted work session.
Engineers coordinating multiple agents on the same project
Multiple agents working on different branches or tasks can signal each other through the socket API. One agent can wait for another to reach a blocked state before reading its output and continuing, enabling genuine asynchronous multi-agent workflows within a single Herdr workspace.
Skip if:
You are running a single agent on a single task. Herdr's coordination API adds value only when agents need to hand off work or check each other's state.
Developers who switch between machines regularly
Add home desktop, work laptop, and a rented server as Herdr machines. All workspaces and agents appear in the same sidebar. Switch physical machines without losing terminal state or restarting sessions.
Skip if:
You work exclusively from one machine and never need remote access. Single-machine users get persistence and agent status monitoring but not the cross-machine workspace view.
Developers building custom agent workflows
The socket API and 1,548-plugin marketplace let you extend Herdr: custom status indicators, webhook integrations, cross-agent prompts, and pane layouts defined in code. The agent skill docs show how to give your CLI first-class Herdr support.
Skip if:
You want a zero-configuration agent runner with no API integration. Herdr's extensibility requires some setup; if you only need a single agent to run unattended, a simpler path is less overhead.
The problem it solves#
Coding agents tied to open terminal windows do not survive machine sleep or network disconnects. A long-running task dies the moment a laptop lid closes, a VPN drops, or you switch from your desk machine to your laptop. Running multiple agents across multiple projects compounds the problem: you end up with a pile of terminal windows with no way to see which agent is blocked, which is done, and which needs a response from you.
The result is manual babysitting. Developers check each window in turn, lose context on what each agent was doing, and restart failed sessions from scratch. Multi-machine workflows require separate sessions on each machine with no shared view. Work that should be asynchronous and continuous becomes synchronized to the human's physical presence at a keyboard.
How it solves it#
Persistent terminal server
The server runs in the background and owns all terminal processes. Closing the client or losing the network connection does not kill agent sessions. After a server or machine restart, Herdr restores the saved layout and can resume supported agent sessions; the original processes do not survive a hard restart.
Agent status visibility
Every pane is marked working, blocked, or idle in real time. When an agent stops and needs input, it surfaces in the agent list without you scanning each terminal manually. A single sidebar shows all agents across all workspaces and machines with their current state.
Multi-machine workspaces
Add local machines and SSH remotes; their workspaces and agent lists appear alongside each other in one client. Switch machines without opening another client connection, and agents on remote machines keep running during the transition. Herdr Cloud (in development) will remove the SSH setup requirement.
Agent-native CLI and socket API
Agents can split panes, start other agents, prompt each other, and wait until another agent is genuinely blocked before continuing. The same surface the CLI exposes to humans is the surface agents use, which means any CLI action is scriptable. The socket API is documented for agent developers building Herdr support.
Works with 22 agent CLIs out of the box
Claude Code, Codex, Cursor, opencode, Grok, GitHub Copilot, and 16 more are detected and supported without configuration. Herdr does not wrap or replace them; it owns their terminals. Any agent CLI that runs in a terminal works, including those not in the default detection list.
Plugin marketplace
1,548 community plugins extend pane behavior and workflows. Plugin authors publish to the public marketplace; users browse and install from the client. Plugins can add custom status indicators, workflow integrations, and layout presets. See herdr.dev/plugins for the full catalog.
Strengths and trade-offs#
Strengths
- Apache 2.0 license with full self-hostingRun on your own hardware with no usage fees, seat limits, or data leaving your network. The Apache 2.0 license permits commercial use and modification without restriction, with no additional terms beyond the standard license text.
- Single Rust binary, no ElectronThe entire client is a single compiled Rust binary that runs in your existing terminal emulator. No separate desktop application, no embedded Chromium, and no background renderer process. Memory footprint is smaller than comparable Electron-based developer tools.
- Agent-native coordination APIThe socket API gives agents a genuine coordination layer: spawn panes, read pane state, signal completion, and wait for another agent's blocked state. This is a first-class protocol, not keystroke injection. Agent CLIs can build Herdr support directly using the published API.
- 22 agent CLIs supported out of the boxClaude Code, Codex, Cursor, opencode, Grok, GitHub Copilot, and 16 more are detected and configured without manual setup. Each gets its own pane with state tracking. Adding support for a new CLI is documented in the public API for agent developers building integrations.
Trade-offs
- -Herdr Cloud is not yet availableThe multi-machine feature currently requires configuring SSH on each remote machine. The Herdr Cloud option for connecting machines without SSH setup is on the waitlist and has no announced launch date. Teams unfamiliar with SSH will need setup time before using cross-machine workspaces.
- -333 open issues at rapid growth stageThe project had 43,271 GitHub stars and 333 open issues as of October 2026, seven months after launch. The active development pace means the issue tracker reflects a proportionally large user base. Bugs that appear after each release tend to be addressed quickly but volume is high.
- -Session state recovery has limitsAfter a server or machine restart, Herdr restores the saved layout and can resume supported agent sessions, but the original processes do not survive. Agents that do not support the Herdr resume protocol restart from scratch. Long-running tasks that have not checkpointed will need to be re-run.
Herdr vs alternatives#
Herdr vs GitHub Copilot Workspace
Both tools let developers hand off coding tasks to an AI agent and check back on results. GitHub Copilot Workspace is a cloud-hosted service where agents run in GitHub's infrastructure. Herdr is a self-hosted runtime that runs agents on machines you own.
| Feature | Herdr | GitHub Copilot Workspace |
|---|---|---|
| License | Apache-2.0 | Proprietary |
| Hosting | Self-hosted (your hardware or server) | GitHub-managed cloud only |
| Agent CLIs supported | 22 (Claude Code, Codex, Cursor, opencode, and more) | GitHub Copilot only |
| Session persistence | Background server keeps sessions alive | Cloud-managed |
| Multi-machine | SSH remotes and local in one view | Single cloud environment |
| Pricing | Free (self-hosted) | Paid (GitHub Copilot subscription) |
Herdr is the better choice when you want to run existing agent CLIs with session persistence and multi-machine access, without paying per-seat fees or sending code through a third-party server. GitHub Copilot Workspace is worth considering when your workflow is entirely within GitHub's ecosystem and you want a zero-infrastructure setup with Copilot as the only agent.
Herdr vs Cursor Background Agents
Cursor's background agents run tasks in cloud sandboxes. Like Copilot Workspace, they are managed and cloud-only. Herdr runs on your own hardware and supports Cursor itself as one of the 22 agent CLIs it manages.
| Feature | Herdr | Cursor Background Agents |
|---|---|---|
| License | Apache-2.0 | Proprietary |
| Hosting | Self-hosted | Cloud sandboxes |
| Supported agents | Any of 22 CLIs | Cursor only |
| Data residency | Your infrastructure | Cursor's cloud |
| Pricing | Free (self-hosted) | Cursor subscription required |
Herdr is the better fit when you need data to stay on your own infrastructure, want to coordinate multiple different agent CLIs in the same workspace, or need session state to survive across machine transitions. Cursor's background agents are the simpler choice when you already live in the Cursor IDE and do not need to run other agent CLIs alongside it.
Quick start#
Install Herdr using the setup script on macOS or Linux.
```bash
curl -fsSL https://herdr.dev/install.sh | sh
```What it's built on#
- Languages
- PythonRustTypeScript
FAQ#
Is Herdr free to use?
Yes. Herdr is Apache 2.0 licensed and free to self-host on any machine. The project has no paid tiers for self-hosted use. A Herdr Cloud feature for connecting machines without SSH setup is in development and its pricing has not been announced.
Does Herdr work with Claude Code?
Yes. Claude Code is one of 22 agent CLIs Herdr detects and supports out of the box. Herdr owns the terminal Claude Code runs in without wrapping or replacing it, so all Claude Code features continue to work as normal. The socket API lets Claude Code agents read pane state and coordinate with other agents.
What happens to agent sessions when the machine restarts?
Herdr restores the saved terminal layout after a server or machine restart and can resume supported agent sessions. The original processes do not survive the restart, so agents that do not support the resume protocol start fresh. See herdr.dev/docs/session-state/ for which agents support resume.
Can Herdr run agents on a remote server?
Yes. You can add remote machines over SSH and their workspaces appear alongside local ones in the Herdr client. Agents on the remote machine run and persist independently of whether you are connected. Herdr Cloud, which removes the SSH configuration step, is on the waitlist.
How does Herdr compare to tmux?
tmux is a terminal multiplexer that keeps sessions alive and splits panes, but it has no awareness of what is running inside those panes. Herdr adds agent state detection (working, blocked, idle), a socket API for inter-agent coordination, multi-machine workspaces, and native support for 22 agent CLIs. The two can coexist; Herdr is not a drop-in replacement for tmux but serves a different purpose.
Similar open-source tools#
ax
Declarative runtime for autonomous agent workloads
FckSignups
Open-source tools that work instantly, no signup required
agent-toolkit-for-aws
Empower AI agents to build and manage AWS applications
Omnara
Open-source agent deployment API. Self-host or use Omnara Cloud.
CodeBoarding
Review code changes as architecture maps, not line diffs
Hypa
Local-first terminal workspace for AI coding agents