We build software with AI employees. Then we check it.
kaimoi grew out of the way we run our own work: two AI agent roles that plan, build and review, and a human who signs off. We turned it into a product so anyone with a Claude Code or Codex session can run the same team.
Why we built it
AI agents write code fast, but a single agent grades its own homework. It says a task is done when it isn't, or a commit exists when it doesn't. We wanted the speed without the blind trust, so we split the work: one role owns the plan and the review, the other owns the build.
How we work
Kai leads: takes the task, plans it, writes the tests that define "done" and reviews every change. Moi builds: writes the code and makes the tests pass. A human decides: nothing important ships without a person looking at it.
We use it every day
kaimoi is built with kaimoi. Kai and Moi pick up our tasks, run our automations and keep our wiki up to date. When they get something wrong, we write about it in our newsletter, Three Hands on the Work, and fix the process.
A few principles we don't bend.
Your agent, your tokens
Agents run on your own machine with your own subscription. We don't resell model access or run your code on our servers.
Verified, not assumed
Every piece of work is checked by a second role against tests and evidence before it's reported as done.
A human has the last word
Responsibility isn't a capability. Someone has to stand behind the work, and that someone is a person.
Honest records
Tasks, reviews, logs and token usage are kept in one place, so you can see what really happened.
Want to work with kai & moi?
kaimoi is in beta and opens by invitation. Tell us about your project and we'll get you set up.