Herd your coding agents.
A Rust CLI and TUI that groups Claude Code agents into herds, keeps them working across restarts, and wakes you only when one needs you.
v0.1.0 · GPL-3.0 · Linux · early (0.x)
## the problem
Five Claude Code sessions, three repos, a dozen tabs. One's waiting on a question, one finished an hour ago, one died with the last reboot. loopherd keeps track so you don't have to: it knows what each agent is doing, brings them back after a restart, and tells you the moment one needs you.
## what it does
- Templates
- Declare once how a member spawns (repo, worktree, prompt), then add one with a single command.
$ loopherd add my-project pr 1050 - One view of every herd
- Members across repos, sorted by what needs you.
$ loopherd view - Survives restarts
- A ledger of each member's task and Claude session. After a reboot or a closed tab, bring them all back.
$ loopherd revive - Queues
- loopherd feed fills free slots from any command that prints tasks, and retires members whose work is done.
$ loopherd feed my-project - Mailbox & watch
- Members report from their own pane. loopherd watch blocks until one needs you.
$ loopherd watch - Leads
- A lead member runs the herd's feed/watch loop for you. It can only merge, install or rebase once you turn that switch on, and a hook enforces it.
$ loopherd lead my-project
$ loopherd view

## requirements
- - jq
- - herdr
- - Claude Code
! Linux only.
## install
$ curl -fsSL https://junderdo.dev/loopherd/install.sh | sh # It installs the latest release as ~/.local/bin/loopherd and adds hooks to your Claude Code settings.json, backing it up first. Read the script before you pipe it to sh.
# Pin a release with LOOPHERD_VERSION=v0.1.0 in front of sh.
# Tip: trust each repo once from its main checkout (cd <repo> && claude, accept, exit) before spawning there.
update: run the same line again
uninstall: loopherd install-hook --remove and rm ~/.local/bin/loopherd
from source (needs Rust): git clone https://github.com/junderdo/loopherd && cd loopherd && scripts/install
## first herd
# ~/.config/loopherd/herds/my-app.toml
repo = "~/code/my-app"
[template.pr]
params = ["number"]
name = "pr-{number}"
worktree = "pr-{number}"
prompt = "You're {name}. Check out PR {number}, read its review and CI, and report what's needed." $ loopherd add my-app pr 42 $ loopherd status $ loopherd view
Feed, leads and handoff are in the README tour →
## how it's built
loopherd is built largely by its own herds: agents pull cards, and tests and lints push back. The working agreement they follow starts at AGENTS.md →