Open Source
Tell the AI what to do.
It ships the rest.
Set your repo up once, then direct Claude Code in plain language — from your desk or your phone. It researches, builds, and opens a pull request; an AI reviewer gates every change; and the agent fixes its own review notes until it merges. You point, it ships.
Open source · runs in your own GitHub · built for Claude Code
Direct, verify, done
The same loop we use to build human0 itself — including this page. Two touchpoints are yours: say what you want, and say “go.” The rest runs itself.
You say what you want
Describe the change in plain language — a bug fix, a feature, a doc update. Kick it off from your laptop or your phone with Claude Code on the web. No tickets, no ceremony.
The agent builds the whole thing
It researches the code, makes the change, updates the docs, and keeps the repo organized — following the conventions you wrote down once. Then it opens a pull request with a live preview.
You glance and say “go”
Check the preview to make sure it's what you wanted. One word clears it to ship — and that's the last time it needs you.
The reviewer gates it — and the agent clears it
An AI reviewer checks every change like a careful senior engineer: correctness, security, tests, your rules. When it asks for changes, the agent fixes them and asks for another pass — looping on its own until both sides agree.
It merges itself
Approved and green, the change lands. No button to click, no thread to chase. You directed it and verified it — you didn't babysit it.
The secret is a file your AI actually reads
You write your conventions, your structure, and your standards once — in
a plain-language file called AGENTS.md (and its twin
CLAUDE.md). Every agent reads it on every run, so it works
the way your team works without being reminded each time.
The AI reviewer reads it too — so the same standards you'd enforce by hand are the ones it holds the line on. That's what makes it safe to let the AI ship: nothing merges that doesn't clear your bar.
Why teams set it up
Built for Claude Code
Claude Code is the agent we build on — including Claude Code on the web, so you can start a task from anywhere and it runs in the cloud. Codex or any other agent in your repo works too.
It clears its own review
When the reviewer asks for changes, the agent reads the feedback, fixes it, and re-submits — looping until it's approved. The cycle closes without you in it.
A reviewer that holds the bar
Every change gets a thorough, consistent review before it merges. Letting the AI ship doesn't mean lowering quality — it means the bar is enforced every single time.
You stay in control
Verify the result before it ships, or let it merge and check the history. Pause it, change its rules, or take the wheel anytime.
Your code stays yours
It runs inside your own GitHub, on your own account. Nothing is copied out, stored, or shared.
Open source, set up in minutes
Apache 2.0 — fork it, read every line, make it yours. Add one key, drop in a couple of files, and the loop is working on your very next change.
We trust it with our own company
Most of what human0 ships is built by AI. This very website, the articles, most changes — directed in plain language, built by an agent, reviewed by this setup, and merged on their own once they passed. It's not a demo. It's the loop everything we ship runs through.
200
Changes shipped
2.8h
Median time to merge
99%
Of changes reviewed by AI
23/24
Hours of the day that shipped
Counted from our own repository, 10 July – 9 August 2026. Read the full case study.
Two ways to get started
Add it to a project you already have, or start something new from the template.
Add it to your repo
Already have a project? Add one key and one workflow file, and the AI reviewer runs on your next pull request. Nothing else changes about how you work.
Get the reviewerStart from the template
Starting fresh? Fork the template and your new repo comes with the guidelines agents read, the reviewer, and a workflow built for shipping by instruction — all wired up.
Use the templateComing soon
This is just the first layer
Directing one repo is where it starts. Next comes the platform: a system for steering a whole team of AI agents — building, planning, reviewing, and running operations — with you in the loop exactly where it matters. The full vision is a company that runs itself.
It's still in development, and we're looking for a few early candidates to shape it with us.