One command. Parallel agents.
One reviewed PR.
HerdOS orchestrates AI coding agents through GitHub. Describe a feature, walk away, and come back to a reviewed pull request ready to merge.
You're not building. You're babysitting.
You open four terminal tabs. You assign each agent a task. You check back every ten minutes to see if anyone's stuck. You're not building — you're babysitting.
Two agents touch the same file. You get a merge conflict. Now you're the human merge resolution layer, untangling AI-generated code you didn't write.
An agent stalls silently. No error, no notification. You discover it an hour later, kill it, and start over. That's an hour of compute and context — gone.
Coordinating AI agents shouldn't be your full-time job.
What the system handles for you
You describe a feature. The system handles everything else.
Built for the AI-native workflow
Parallel Execution
Multiple agents work simultaneously on isolated branches. No contention, no waiting — just throughput.
GitHub-Native
Built on Issues, Actions, and PRs. No custom infrastructure, no databases, no daemons to run.
Self-Healing
Crashed workers are detected and redispatched automatically with exponential backoff. No silent failures.
Conflict Resolution
When parallel agents touch overlapping code, merge conflicts are detected and resolved without human intervention.
Agent Review
An AI reviewer checks the consolidated PR for bugs, security issues, and style — before you ever see it.
Multi-Provider
Claude Code, OpenCode, and Codex are first-class today — pick by capability, billing model, or both. Mix providers per role: Claude for planning, Codex for workers. Gemini CLI and Cursor are on the roadmap.
You stay in control
Workers run autonomously, but nothing merges without your approval. You review the PR, you decide what ships.
Zero Infrastructure
No database to run, no daemon to keep alive, no local compute. HerdOS shuts down completely when it's not working. Your laptop battery thanks you.
Who this is for
Solo developers using AI coding agents who want to parallelize work without managing multiple terminals
Small teams coordinating agents across a shared codebase
Anyone already on GitHub who wants agent orchestration without new infrastructure to deploy or maintain
HerdOS requires self-hosted GitHub Actions runners. If you're not on GitHub or can't run self-hosted runners, it's not the right tool yet.
Built for real features
herd plan "Add user authentication" herd plan "Add Stripe billing and subscription management" herd plan "Add dark mode support" herd plan "Add end-to-end test coverage for the checkout flow" herd plan "Add rate limiting, request logging, and API key management" herd plan "Upgrade React 17 to 19" Inspired by Gastown, Steve Yegge's multi-agent orchestration system. HerdOS carries forward its proven ideas — role decomposition, batch-based delivery, nondeterministic idempotence — but replaces tmux sessions, Dolt databases, and battery-draining polling daemons with GitHub Issues, Actions, and event-driven workflows. Zero infrastructure. Zero local compute.
Quick Start
Up and running in minutes
Also on AUR (yay -S herd-bin), pre-built binaries, or from source — see all install methods.
Note: HerdOS runs workers on self-hosted GitHub Actions runners. You'll need at least one runner set up before dispatching. See the runner setup guide →
Ready to herd your agents?
Ship features faster with parallel AI agents — no babysitting required
Get Started on GitHub