Kilo AI

Kilo, also known as Kilo Code, is an open-source AI coding agent. It works in VS Code and JetBrains editors, in the terminal through a CLI, in the cloud and in a beta desktop app for macOS, Windows and Linux. The idea is to have one open agent across every place a developer works.

Models and keys

Kilo gives access to more than 500 models, including frontier models and open-weight models. The company says it charges zero markup on model usage. You can also bring your own API keys or run local models, which can help with privacy when code should not leave your machine.

What the agent does

Like other coding agents, Kilo can read files in your project, edit code, run commands such as tests and explain what it is doing. The site shows an example where the agent investigates a flaky test, finds a race condition, changes the code and reruns the test 50 times to check the fix. The desktop app lets you run several agents side by side in one window, for both code and data tasks.

Pricing

Individuals can start free with the open-source extensions for VS Code, JetBrains and the CLI; model usage is billed separately. Teams plans add collaboration features, and Enterprise adds governance controls for larger organisations. Check the pricing page for current details.

Who it suits

Kilo suits developers who want an open-source coding agent, choice over which models they use, and the option to work in their editor or terminal. Teams that want to control costs by paying model providers' rates directly may also find the zero-markup approach useful.

Things to consider

Because usage is billed per model, costs depend heavily on which model you choose and how much work you give the agent; set spending limits early. Review every change the agent makes, run your tests and use version control so you can undo edits easily. Be careful with commands that touch production systems or secrets, and check the permission settings that control what the agent can run without asking. If privacy matters, confirm which data is sent to model providers or use a local model, and read the team settings before rolling it out widely.