Herdr Projects lets you run a larger piece of work in Herdr when one agent isn't enough and managing five by hand is a job in itself, by giving you one coordinator conversation that starts a separate agent for each task on its own branch, gives every one of them the same instructions and memory, and shows you which threads are ready for review, waiting on you, or still working.

_{✓ Free, MIT licensed   ✓ Runs on your machines, no hosted service   ✓ macOS and Linux, Herdr 0.9.1+}

The coordinator never does the work itself, so it's always free to answer you. Each task runs in its own thread: a separate agent in its own git worktree and branch, or in its own folder when there's no repository. You read reports and answer the threads that need you instead of briefing every agent yourself.

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

Each thread shows its project, its id and its group beside its session: ready-for-review, waiting-on-you, working, landing or idle. One action filters the sidebar to a single project, with the coordinator first and finished work next, and a text popup prints the same groups.

Say what you want once. The coordinator proposes threads and waits for your go-ahead, then each thread starts from a brief with your standing instructions, the project's memory and its task. Lessons a thread reports under ## Remember flow back into memory for the next one.

A thread that waits on a permission prompt for more than 30 seconds moves to Waiting on you, and a finished one stays under Ready for review until you've looked. A background ticker follows pull requests, runs your scheduled routines, and leaves each change in the coordinator's inbox.

For developers who run coding agents in Herdr on macOS or Linux Free    ✓  A coordinator that delegates and never does the work itself    ✓  Threads on their own worktree and branch, or in a tab    ✓  Shared instructions and memory in every brief    ✓  Overview by what needs you, in the sidebar and as text    ✓  Pull request follow-up, routines and watched commands    ✓  Threads on your saved SSH machines, reports copied home

You need to be comfortable in a terminal. The plugin builds itself on install, and a project is a plain folder of Markdown and TOML files. You'll need macOS or Linux, Herdr 0.9.1 or newer, Rust/Cargo, Git, and an agent CLI Herdr can start, such as Claude Code. The getting-started guide covers the prerequisites.

Run:

herdr plugin action invoke doctor --plugin herdr-projects

The command checks the Herdr version, the tools it calls, the ticker, and each project's session. The getting-started guide walks through a project that won't open, a thread that doesn't start, and a ticker that isn't running.

It runs the herdr-projects binary every turn, so you'll want to allow-list it in your agent by subcommand, never the bare binary. Allow reading and steering (skill, context, inbox done, thread list, thread prompt and the like) and leave thread resolve, delete, routine approve and open on your agent's normal permission prompt. Leave thread start off the list too unless you've set start_threads = "auto": then every thread start is a real confirmation. Operations has the exact patterns.

No. A project is a folder on your machine, the plugin talks to your local Herdr session, and remote threads use your own SSH machines. Your agent CLI still uses its own service as usual.

No. It never removes a worktree, deletes a branch, merges or pushes on its own. thread resolve --remove-worktree removes a worktree only when you ask, never with force, and only after the thread's report and files are copied home. Deleting a project moves its folder to a trash folder and leaves every worktree and branch alone.

In ~/.herdr-projects/<name>/ by default: PROJECT.md for your settings and standing instructions, MEMORY.md and memory/ for what the coordinator remembers, TASKS.md for the task list the coordinator keeps for you, threads/ for thread records and reports, and library/ for files threads produced. The safety settings live outside it, in ~/.config/herdr-projects/config.toml, where no agent works. Operations lists every file.

No. Run herdr-projects thread start yourself, or take an agent pane you already started and make it a thread with thread adopt. The Projects: continue this workspace as a project action turns the workspace you're in into a project with its agent as the first thread.

They are soft. By default the coordinator proposes threads and waits, thread agents keep their normal permission prompts, and routines may not run shell commands until you enable them and approve each command in a terminal. But agents have a shell: one that runs with skip-permission arguments can edit those files, a thread can prompt the coordinator pretending to be you, an approved routine command covers its text and not the scripts it calls, and whatever reaches memory is repeated in every later brief. Operations says plainly what each guard stops and what it doesn't.

Herdr Projects is free and MIT licensed. You need access to the private repository to install it. Your agent CLI's usual usage charges still apply: every thread is a full agent session, and the coordinator spends tokens each turn reading its digest.

Your first project starts with an install, a name, and one sentence about what you want. Herdr Projects starts the threads and keeps them in view. You choose what to review and what to merge.

_{✓ Free, MIT licensed   ✓ Runs on your machines, no hosted service   ✓ macOS and Linux, Herdr 0.9.1+}