What is herdr?
herdr calls itself the runtime your coding agents live on. In practice it is a terminal workspace manager in the tmux mould: one Rust binary runs a background server that owns every terminal pane, and a client attaches to it, detaches with ctrl+b q and reattaches with herdr. Close the laptop or lose the SSH connection and the agents keep working. It does not wrap or replace the agents; Claude Code, Codex, Cursor, OpenCode, Grok and the rest run unchanged in real terminals, and herdr reads each pane to mark them working, blocked, done or idle.
Around that server it adds workspaces, tabs and split panes, Git worktree checkouts opened from the sidebar, saved SSH machines shown in the same window, plugins, and a CLI and socket API that agents themselves use to start, prompt and wait on other agents. After a server restart it restores the layout and can resume supported agents' own sessions, though the original processes do not survive.
The project is from Herdr, Inc., licensed Apache-2.0, created in March 2026, and installs with a shell script, Homebrew, mise or a PowerShell script on Windows. It has about 47,100 stars and 3,760 forks; the latest stable release is v0.9.5 (14 October 2026). Since October the same team also ships a separate desktop app built on a new herdr-core server; that app does not share agents with this terminal version, which herdr.dev now calls Herdr TUI.
Is herdr worth it?
herdr is the strongest open answer yet to running many terminal agents at once, and the reason is that it manages terminals rather than agents. Here "runtime" means a background server that owns real panes, not a sandbox, a container or a wrapper: the agent CLI runs exactly as it would in any terminal, and herdr watches it. It reads the live bottom of each pane against a detection manifest for each agent, marks it working, blocked, done or idle, and rolls that state up to the tab and workspace, so the one stuck on a permission prompt shows in the sidebar instead of being found by cycling through windows. The site counts 22 agent CLIs detected out of the box.
The persistence model is spelled out more carefully than most. Detach and every process keeps running. After a full server restart, herdr restores workspaces, tabs, panes and working folders, keeps up to 48 layout snapshots, and resumes the agent's own conversation for the 18 integrations whose resume command it knows, such as claude --resume and codex resume; anything else comes back as a fresh shell in the right folder. Pane screen history is off by default because terminal output can hold secrets. Saved SSH machines appear in the same sidebar over your normal OpenSSH authentication, and the socket API lets one agent split a pane, start a reviewer and wait until it is genuinely blocked.
Against the neighbours it sits in a clear place. Plain tmux gives you the same detach and prefix keys but knows nothing about agent state. Claude Squad builds a smaller version of this idea on top of tmux, one worktree per session; herdr owns the terminals itself and offers worktrees as an option from the sidebar rather than the whole model. T3 Code goes the other way, a graphical and phone surface that drives agent CLIs directly. herdr stays in the terminal; phone access comes from a client such as Ovis or from SSH.
The catch is that herdr adds reach and visibility, not isolation. Agents run with your user's full permissions, plugins are neither reviewed nor sandboxed, and the docs warn that an idle reading is not authorisation to approve anything, since an unfamiliar prompt can show as idle until the manifest learns it; one open issue reports exactly that for a Claude Code permission dialog. Detection rules update from herdr.dev in the background unless you turn that off. Pace is the other cost: about 415 open issues and pull requests, stable releases every week or two plus preview builds, and a second desktop product whose agents do not show up here, which makes the project's direction worth watching before you standardise a team on it.
Where it earns its place
A backend engineer running Claude Code and Codex on a rented Linux server
Keep long agent runs and a six-hour backfill going while the laptop sleeps
Install herdr on the server and attach with herdr --remote over your existing SSH config; panes keep running server-side and reconnect on return. The tool is free; the bill is the agents' own plans plus the server.
A senior developer juggling four repositories with an agent in each
Stop checking every pane to find the agent stuck on a permission prompt
One workspace per repo, worktrees from the sidebar for parallel branches, and the sidebar's blocked marker points at the pane waiting on you. Parallel agents multiply token use, so plan limits arrive sooner.
A small platform team building internal agent tooling
Let one agent hand a review to another and pick up its report without copy and paste
Use the CLI or socket API to spawn a pane, prompt a second agent and wait until it is blocked or done; Apache-2.0 allows building closed internal tools on it. The cost is engineering time against an API still at v0.9.
Before you commit
- Setup
- curl -fsSL https://herdr.dev/install.sh | sh, brew install herdr or mise use -g herdr (PowerShell script on Windows), then run herdr in the project folder. Optional herdr integration install <agent> adds state reporting and session resume per agent.
- Integrations
- Detects Claude Code, Codex, GitHub Copilot CLI, Cursor Agent, OpenCode, Pi, Droid, Devin, Kimi, Qwen, Grok, Antigravity, Amp, Gemini CLI and others; a CLI and Unix socket API for scripts and agents; a plugin system; saved SSH machines.
- Data and privacy
- Runs locally; agents talk to their own model providers on their own terms. herdr fetches detection-manifest updates from herdr.dev unless [update] manifest_check = false, and remote attach can download binaries from herdr.dev. Session files hold layout and agent session references; pane history is opt-in. Apache-2.0; no security advisories published.
- Cost at scale
- Free for any team size. Cost is the agent subscriptions or API usage, which rises with each parallel pane, plus any servers you attach over SSH. Herdr, Inc. lists an enterprise contact but no paid plan.
- Limits
- No isolation beyond your own user; screen-based state detection can misread unfamiliar prompts; recurring issue themes are Windows terminals, keyboard protocols in Ghostty and image rendering; live handoff on update is experimental; agents in the new Herdr desktop app are not visible here.
What’s good
- Marks each agent working, blocked, done or idle and rolls it up the sidebar
- Agents keep running after detach, SSH drops or closing the client
- Resumes 18 agents' own sessions after a server restart
- Socket API lets agents start, prompt and wait on other agents
- Single Rust binary for macOS, Linux and Windows
What’s not
- No sandbox; agents and plugins run with your full permissions
- Blocked detection reads the screen and can miss new prompts
- About 415 open issues and pull requests, fast release cadence
- Separate desktop app on herdr-core does not share agents with it
Best for
Terminal-first developers running several coding agents across projects or machines who want them to outlive the client and to see at a glance which one needs an answer.
herdr alternatives
herdr: common questions
- Is herdr worth it?
- herdr is tmux rebuilt around the question you actually ask when five agents are running, which one is waiting on me, and it answers it well. We rate it 4.4 out of 5.
- How much does herdr cost?
- Free and open source (Apache-2.0). You bring your own agent subscriptions or API keys; no paid tier is published.
- What are the downsides of herdr?
- No sandbox; agents and plugins run with your full permissions; blocked detection reads the screen and can miss new prompts; about 415 open issues and pull requests, fast release cadence; separate desktop app on herdr-core does not share agents with it.
- Who is herdr best for?
- Terminal-first developers running several coding agents across projects or machines who want them to outlive the client and to see at a glance which one needs an answer.
- How much does herdr cost for a team or in production?
- Free for any team size. Cost is the agent subscriptions or API usage, which rises with each parallel pane, plus any servers you attach over SSH. Herdr, Inc. lists an enterprise contact but no paid plan.
- Is herdr safe for company data?
- Runs locally; agents talk to their own model providers on their own terms. herdr fetches detection-manifest updates from herdr.dev unless [update] manifest_check = false, and remote attach can download binaries from herdr.dev. Session files hold layout and agent session references; pane history is opt-in. Apache-2.0; no security advisories published.
- What are the best herdr alternatives?
- The closest alternatives we have reviewed are Claude Squad, T3 Code and Ovis.
Answers checked Oct 31, 2026.
Make herdr? Show this rating on your site
