AI Coding Agent for Real-World Software Projects

Kimi Code is an AI coding agent that reads your codebase, edits files, runs commands, tests changes, and turns natural-language requests into reviewable code. You define the goal, review the plan, approve important steps, and inspect the final diff from the terminal, browser, IDE, or on the desktop.

From context to reviewable code

Understand any codebase, starting from zero

Open a repository and ask Kimi Code how it works. Starting at the entry point, it traces the execution flow and maps how the modules connect, producing a structured walkthrough of the codebase.

Turn multimodal contexts into actionable tasks

Programming work goes beyond code and written requirements. Kimi Code can read multimodal inputs such as log screenshots, design references, charts, and even videos. It pulls out the key details and brings them into the task as working context.

Turn requirements into reviewable changes

Describe the goal in plain language, and Kimi Code works out where it belongs in the codebase. It identifies the implementation path, updates the necessary files, and returns a focused diff instead of a wall of edits.

Deliver code changes that already pass tests

Kimi Code runs the test suite, reads the failures, and fixes what breaks, iterating against real feedback from the development environment. By the time a change reaches review, it has already been exercised.

Kimi's AI coding agent, make it work the way you work

1

Turn good workflows into skills

Package proven workflows as skills. Kimi Code follows the same process every time to standardize code reviews, task execution, content creation, and more.

2

Sync your habits with hooks

Hooks run predefined actions at key moments in a task, from enforcing team conventions to triggering internal scripts, so the agent works the way you do.

3

Call familiar tools through MCP

Connect Kimi Code to tools you already know. Tools with MCP server can be called mid-task, so the agent works with your real stack.

4

Package stronger capabilities as plugins

Bundle skills, hooks, & MCP configuration into one plugin. Give Kimi Code a complete working setup, ready to share across the team.

Coding agent that turns good workflows into skills

Built for long-horizon tasks

AI programming agent that plan first and then execute

Explore before editing & execution

For complex changes, Kimi Code can use plan mode to understand the repository before making edits: it explores the relevant files, identifies the likely implementation path, and turns the work into a clear plan.

AI code agent that keeps moving toward a goal

Keep moving toward a goal

With “/goal”, Kimi Code can work toward a defined outcome. Given an objective and completion criteria, it tracks progress, chooses the next useful step, and continues until the goal is complete.

AI coding agent that can split the work and then merge the results

Split the work, merge the results

Kimi Code can split work across sub-agents with independent contexts. For batch tasks, Agent Swarm can launch multiple sub-agents at once, aggregate their findings into the main workflow, and improve overall efficiency.

AI code agent that moves faster, stays thoughtful

Move faster, stay thoughtful

Kimi Code high-speed model delivers output 5–6× faster than the standard edition. Switch to it for routine work and get the same coding capability with far less waiting. Speed becomes a choice, not a trade-off.

FAQ

An AI coding agent can read and understand your codebase, edit files, execute commands, and run tests to turn natural-language requests into reviewable code changes. Kimi Code brings these capabilities to your terminal, IDE, and desktop app, so you can work wherever you feel most productive.
Code-completion tools suggest the next few lines while you type. Kimi Code is an AI coding assistant that works at the task level: it plans a change, edits multiple files, runs the relevant checks, reads the results, and iterates until the work is ready for your review.
The AI coding agent, Kimi Code, can run as an interactive CLI in your terminal, as a browser UI via kimi web, and inside editors. It also supports the ACP protocol, so editors such as Paseo, JetBrains, and Zed can connect to it through the CLI. Now Kimi Code has a desktop version, available for macOS, Windows, and Linux.
You can install Kimi Code as a desktop app, a terminal CLI, or a VS Code extension. For the desktop app, download the installer for your operating system from the official website. For the terminal CLI, follow the installation guide, then run kimi in your project folder. For VS Code, install the Kimi Code extension from the VS Code Marketplace. Follow the setup prompts to sign in or configure a model, then open your project and start a conversation.
Yes. Kimi Code works with mainstream coding agents such as Roo Code and OpenCode. Create an API key in the Kimi Code console, point the tool to the Kimi Code endpoint, and set the model ID to kimi-for-coding to use your quota there.
Yes. Plan mode lets you review the approach before edits begin, /goal keeps the agent working toward a defined outcome, and Agent Swarm splits large jobs across parallel sub-agents. So an AI development agent can carry a task from requirement to reviewable change without constant supervision.

Your coding agent, from context to commit