xherd
Run a herd of coding agents in parallel: one screen, one ticket per sandbox, and you approve every step.
Project intro
xherd runs a herd of coding agents for you, so you can put several tickets in motion at once without losing track. You describe the work, the agents do it in parallel, and one screen shows every agent, what it is doing, and which ones are waiting on you.
Each task gets its own git worktree and its own tmux session, so agents never step on each other or on your checkout. Agents follow a spec-driven flow, spec, design, tasks, commits, pull request, and stop at approval gates, so you read what they propose before anything moves on.
The web UI keeps it hands-on: a live terminal per agent, diffs, a 'your call' band for whatever needs your decision, and settings with setup checks. Approvals are hashed, so you approve exactly what you read, and agents survive restarts by parking at their gate instead of getting lost.
Two honest limits: it is early and single-user for now, with only one agent working end to end, and it has no stars yet on GitHub. It runs in Docker or natively, with Claude Code today and other runtimes planned, and the code is a single Rust binary with the web UI built in.
Actions
Tech stack
Project info
- Project category
- Developer Tools
- Project status
- Online
- Collaboration intent
- None