herdr-projects

Run dozens of coding agents through one Coordinator without micromanaging each one or losing sight of what needs you

Herdr Projects is mission control for Herdr: one list of tasks, yours and the ones you delegate. You type what needs doing and the board's own agent, the Coordinator, turns it into tasks. Every delegated task runs as an ordinary Herdr agent in its own space and worktree, gets checked and reviewed by a second agent when it finishes, and waits for your yes before it merges. The board shows which tasks need you, which are ready to approve and which are still working.

✓ Free, MIT licensed   ✓ Runs on your machines, no hosted service   ✓ macOS and Linux, Herdr 0.9.3+

Keep one list going while the work happens in parallel

The board is an overview and a remote control; Herdr is the engine. Each delegated task gets its own agent in its own Herdr space, on its own branch in a git worktree, or in a plain folder when there's no repository. Your checkout is never touched. You approve finished work and answer the tasks that need you instead of briefing every agent yourself.

Choose between briefing each agent by hand, one long agent session, a cloud projects product, or a work board in Herdr

Keep your attention on decisions. Herdr Projects starts, checks and tracks the tasks, your agents do the work, and you choose what to answer, approve, or merge.

Tell the Coordinator what you want. See which task needs you on the board.

📈 See every task at a glance

The board's list holds your tasks in five groups, in the order they need you: needs you, ready to approve, working, to do and done. Its header counts them (3 need you · 1 ready · 2 working), and each row says who is on it, where it stands, its review round and pull request, and a percent and time to go. Herdr's own sidebar carries a small copy of each live agent's row, so you see the same thing from any space. With the Herdr web UI the board is on your phone too, and Alt+M shows the tasks of your other machines.

⚡ Stop briefing every agent yourself

Type what needs doing and press Enter: the Coordinator, one long-lived agent chat, makes it a task or several. Send it with Tab instead and every task it adds gets an agent at once. Each agent starts on Claude Code, Codex, pi or another agent herdr supports, with the task and the board's rules, in a worktree of its own. Want more control? Ask for a plan first, several attempts and one kept, limits on time and tokens, or automations that add tasks on a schedule.

💬 Know when a task needs an answer

An agent that asks you something, wants a permission or a sign-in says so to the board, so its task shows under needs you and stays there until you answer. An agent that finishes goes down the finish line: its work is committed, main is merged in, the repo's check runs, a draft pull request opens, and a second agent reviews it. A failed check or review goes straight back to the agent; work that passes waits for your Ctrl+L. With the board closed, a small keeper does the same.

Open the board in three steps

1️⃣ Install the plugin
Run herdr plugin install eliasstravik/herdr-projects. It downloads the one program for your platform: no Node, no npm.

2️⃣ Open the board
Run the plugin's action "Open the work board", or herdr plugin action invoke open --plugin herdr-projects in a Herdr pane. Its first screen offers a key that opens it from anywhere in Herdr.

3️⃣ Type what needs doing
Press Enter and the Coordinator makes it a task, or Tab and each task gets an agent at once. ? shows everything else you can do.

Get everything included, free

For developers who run coding agents in Herdr on macOS or Linux

Free

   ✓  A Coordinator that turns what you type into tasks
   ✓  Every task on its own worktree and branch, in its own Herdr space
   ✓  A check and a second agent's review before anything merges
   ✓  What needs you, on the board, in Herdr's sidebar and on your phone
   ✓  Branches kept up to date with main, cleanup after a merge
   ✓  Plans, attempts, limits, fallback agents and automations

Updating

herdr-projects update           # put the newest release in place of the program
herdr-projects update 1.2.0     # a given version

An open board and the keeper see the new program and load it by themselves, once nothing is half typed and no agent is held up. Pulling a newer plugin with git pull does the same at the next command Herdr runs. A program you built yourself is updated with git pull and cargo build --release, never by a download. Installing has the details.

Get your questions answered

Do I need to know how to code?

You need to be comfortable in a terminal. The plugin downloads one prebuilt program for your platform, and you never have to edit its files: everything changes by asking the Coordinator or from the board's settings page (Ctrl+E). You'll need macOS or Linux, Herdr 0.9.3 or newer, Git, curl or wget, and an agent CLI the board can start: Claude Code, Codex, pi or another agent herdr supports. gh (pull requests) and tailscale (private review links) are optional. Rust/Cargo is needed only when you build it yourself. Installing covers the prerequisites.

How do I check that Herdr Projects is running?

Run:

herdr-projects doctor

It checks the program and its launcher, the plugin in Herdr, your agent programs (installed, signed in) and their model and effort settings, your other machines, free disk space, and leftover worktrees it could clean. Each problem comes with what to do, and it exits non-zero when there is one. Troubleshooting lists every check. Installing walks through the first run, the blocks the board can add to Herdr's config, and taking the plugin out again.

What permissions do the agents get?

Full access by default, so a task never waits on a prompt nobody sees: claude --dangerously-skip-permissions, codex --dangerously-bypass-approvals-and-sandbox. They can run anything as you without asking; the rules and the review are guidance, not a sandbox. Set worker.access ask (and reviewer.access, coordinator.access) to start a program without those flags, so it asks before it acts. Nothing is written to your own agent settings, and your own hooks keep running. How work gets done has the details.

Does the plugin send my work to a hosted service?

No. The board keeps its files in a folder on your machine, talks to your local Herdr session, and reaches your other machines over your own SSH. Reports and review evidence get a private link on your tailnet, or on 127.0.0.1. Your agent CLI still uses its own service as usual.

Will it touch my branches or worktrees on its own?

Only the ones it made. Every task works in its own worktree, on its own branch work/-, never in your checkout. The board merges main into those branches so the reviewer sees the work as it will really land, and rolls a real conflict back at once and hands it to the task's agent. It merges only on your yes, at the commit that was reviewed, unless you set merge.by itself. When a task is done its worktree and branch are removed; a checkout with unsaved or unmerged work stays on disk with its branch. workspace.cleanup false keeps everything.

Where do my tasks live?

In ~/.herdr-projects/ by default: settings/config.json for your settings, data/tasks/// for each task with its closing note and review evidence, data/coordinator/ for the Coordinator's chat, data/automations/ for your automations, and reports/ for reports an agent made for you that belong in no repo. A repo can hold its own .herdr-projects.json with its setup, its check command and a few other facts, but never a setting that loosens a guard. The home folder lists every file.

Can I have several boards at once?

You don't need to: one board per machine holds every repo, each task tagged with its repo, and Alt+S groups them. Each machine has its own board, and Alt+M shows the others' tasks from yours. Each herdr session gets its own Coordinator, but they share the one list. The board key never opens a second board: it goes to the one that runs. Several boards has the details.

Do I have to start every task through the Coordinator?

No. Select a task on the board and press Tab to put an agent on it, or run herdr-projects add "" --delegate and herdr-projects delegate yourself. An automation adds tasks on a schedule or when something happens on the board, such as a merge.

What if the safety settings aren't enough?

They are soft. By default nothing merges without your yes, work whose review failed is never merged by a plain yes, and a repo's own file can't switch on merging by itself, drop the review or change the rules its own branch is checked by. But agents run with full access and have a shell: one can do anything you can, a review is a second agent's judgement and not a proof, and an automation runs with nobody watching. Turn on worker.access ask for prompts, a plan first for a yes before any change, and limits for a cap on time and tokens.

What does it cost?

Herdr Projects is free and MIT licensed. Your agent CLI's usual usage charges still apply: every task is a full agent session plus a review, the Coordinator spends tokens on each message, and the Watcher makes a small model call about once a minute per working agent.

Open the board in three steps

Your board starts with an install, one key, and one sentence about what needs doing. Herdr Projects starts the agents, checks their work and keeps it in view. You choose what to answer and what to merge.

✓ Free, MIT licensed   ✓ Runs on your machines, no hosted service   ✓ macOS and Linux, Herdr 0.9.3+