Kilo Code: open source coding agent for VS Code and CLI

The proposal is simple: you choose the model, switch in the middle of the task and pay the provider rate, with no markup. No API key is needed to start. The account gives access to models like GPT-6 Astra, Claude Fable 5.1, Gemini 3.8 Flash, Grok 4.6 and DeepSeek V4.1 Flash, all at provider price.

In this article you will see:

• What Kilo Code does and where it runs
• How to install on VS Code, terminal and via CLI
• How model selection and switching works
• What the MIT license means in practice
• How it compares to other code agents

What Kilo Code does

Kilo Code is an engineering agent that plans, writes and adjusts code inside your editor or terminal. It reads the repository, proposes changes and runs commands, with you in control of each step.

The difference from traditional extensions is model flexibility. Instead of tying you to a single provider, Kilo Code lets you choose between more than 500 models and switch between them in the middle of a task. One model can plan, another can implement and another can review, without leaving the same flow.

See also how use the Codex CLI in your workflow and how Claude Code 2.1.319 organizes MCP in the background.

How to install

In VS Code, install the Kilo Code extension directly from the Marketplace or via the link vscode:extension/kilocode.Kilo-Code. After creating the account, you already have access to the models.

In the terminal, the fastest path is npm:

$ npm install -g @kilocode/cli

It also works with curl, pnpm or bun:

$ curl -fsSL https://kilo.ai/cli/install | bash

$ pnpm add -g @kilocode/cli

$ bun add -g @kilocode/cli

The main project is at github.com/Kilo-Org/kilocode. New features go to the v2 branch; the main branch receives fixes for the current release. Before contributing, check which branch the team is accepting.

How model selection works

A Kilo Code account gives access to more than 500 models. Billing is the provider rate, with no markup. This changes the equation for people who test models: instead of paying a subscription per tool, you pay for the actual use of each model.

Switching in the middle of a task is the strong point. If a model is good at planning but weak at refactoring, you switch to a second one when writing code. This brings the workflow closer to an agent squad, where each role uses the most suitable model.

What the MIT license means

The repository is MIT. In practice, this means you can use, modify and distribute the code, including in commercial projects, as long as you keep the copyright notice. For teams that need to audit or adapt the tool, the permissive license reduces legal friction.

The combination of open source with direct provider pricing makes Kilo Code an interesting option for those who want control over the agent without giving up top models.

How it compares to other agents

Kilo Code does not try to be the only agent. It competes with Claude Code, Codex CLI and Gemini CLI, among others. The difference is in model flexibility and direct provider pricing.

If you already use a specific agent and are satisfied, there is no reason to switch. If you want to test several models in the same flow without changing tools, Kilo Code is worth a test. The cost of experimenting is low: installation takes minutes and does not require an initial API key.

For teams that already have a flow with Claude Code or Codex, Kilo Code can come in as a comparison tool. Running the same prompt in two agents and comparing the result is a cheap way to calibrate which model performs best in your code.

How to test today

1. Install the extension in VS Code or the CLI in your terminal.
2. Create an account and pick a starting model, such as DeepSeek V4.1 Flash or Gemini 3.8 Flash.
3. Open a small repository and ask for a simple task, like adding a test.
4. Switch models mid task and compare the results.
5. Read the README and open issues before adopting in production.