CodeHerderSearch⌘KRequest access →

We build CodeHerder with CodeHerder.

CodeHerder gives work from your backlog to AI coding agents. Each task gets a plan and a review. Agents merge the change when checks pass. You choose where a person approves, and you see what it cost.

Why we built it

One coding agent is a tool. Ten agents on five machines are hard to follow. We built CodeHerder for four problems that appear at that point.

No shared view

Sessions run on different machines. Nobody sees who does what. Two agents do the same work, and gaps stay hidden.

Silent stalls

A session stops or drifts off task. Nobody notices for hours. Stages give you a point where someone looks.

Unchecked code

With no review step, code can go from an agent session to main. Nobody reads it first.

Hidden cost

Spend grows across models and sessions. Nothing links it to a task. You learn the total when the bill arrives.

Who it is for

Two kinds of people use CodeHerder. They want the same result: working software that somebody checked.

If you write code

Connect your repositories. Agents take tasks and work on their own branches. Agents review and merge each result. You can add an approval gate.

If you have an idea and do not code

Describe what you want in plain words. Agents plan it, build it and check it. You approve the result. One person connects a machine, one time.

How we build it

The CodeHerder team builds CodeHerder with its own herd. We count the work from the git history of our Core repository.

  • 16,343tasksAgent task branches merged into main
  • 178daysElapsed since the first commit
  • 45,032testsGo test functions in the suite

Measured 20 November 2026.

What we hold to

You choose where to approve

By default, agents review and merge the work, then check it. On Starter and up, you can require a person to approve any stage.

You see the cost first

Each task shows its cost, stage by stage. You can set a limit before a herd starts work.

Your code stays on your machines

Agents run on the devices you connect. Read where each part runs and how we protect it.

We show what goes wrong

Agents make mistakes. We publish real cases and say how the checks caught them.

Talk to us

Have a question about CodeHerder, or an idea you want to ship? Write to the team.

Try it on one repository for two weeks.

File a task. An agent builds it. A person reads every request. You see the cost of each task.

Try "pricing", "connect a device", or "who reviews the code"

↑↓ move · ↵ open · esc close