Your coding agent can now run an autonomous engineer

The problem with autonomous agents is not bound by the model anymore. It is about what work gets done and how we can determine when things are truly done.

Until now, you were probably babysitting your coding agent and had to keep track of tasks and features you wanted it to do. You were probably also the one who tested, found any issues and reprompted until late in the afternoon.

You don't have to do that anymore. We have created a new agent that does all of this on its own and even suggests what you should be working on next - without you needing to prompt it constantly.

Free

Kairos is free. Sign in at backant.io, run npx backant-kairos, done. There is no tier to buy and no card to enter. Kairos works on top of your existing coding harness and works with your current AI subscription.

One skill, four agents

backant skill install --all

The skill allows you to use your existing coding agent to work with backant kairos and you never have to read any docs ever again.

This is not just some documentation that you have to read and understand. The skill allows your agent to be able to fully configure Kairos to work alongside you without you ever needing to know any CLI or configuration.

Start simple

Open your agent inside the target repository and say: "Set up backant-kairos for this repo."

It will install the backant-kairos and ask you a few questions and proceed to configure it. Few examples:

May Kairos merge its own pull requests without you clicking merge? Default is no. With it on, PRs that pass CI and its own review go in; a gate can still hold each one for you.

What is the most one turn may cost? Default is 15 dollars.

Which actions must always wait for your yes? Default: opening and merging a PR.

You answer or choose the defaults. The agent configures .backant.toml and will proceed until you have a working configuration. The decisions you made get saved in a file and when you change your mind you just prompt your agent to change it. The changes take effect as soon as kairos finishes its current task.

Give it work

Now you have installed and configured Kairos and it already starts working and familiarizing itself with your repository. As you might have noticed there is no way to chat with Kairos and it is intentional. You usually just let it work. If you do notice something and you want to give it the task just ask your AI agent:

"The integration test on main is flaky. Can you let Kairos take care of it?"

Your agent files a GitHub issue labeled kairos-task and pins it. Kairos takes it on its next turn, opens the PR, watches CI, and if the deploy goes wrong it deals with that too. Ask "what is Kairos doing?" and the agent reads the decision back to you, with the reason that was recorded at the time it was made.

No need to ever talk to it directly. You talk to the agent you were already talking to, and it manages Kairos. Think of it as an additional employee working alongside you.

What is underneath

The engineering itself is currently a claude code instance with memory and a loop. Your agent does not have to be Claude. Codex users, pi users, and Hermes users can continue running their own harness and can control kairos via the skill. Currently the only agent harness underneath is claude code - this is intentional because in our own experience it is less prone to attacks like prompt injection. For an autonomous agent this is a must have in our opinion as you are not babysitting it all the time. We are working on the ability to be able to work with other subscriptions as well though.

How do we know you can trust it: 1-cal.com was built entirely by Kairos, 118 merged PRs, 16 epics, 80K LOC and roughly 5% human intervention.

Start

npx backant-kairos
backant skill install --all

Then tell your agent to set up Kairos. The guide is at docs.backant.io/kairos/agents.