What happens when a Chinese AI lab with a top-ranked open-weight model decides to take on Claude Code, Codex CLI, and Gemini Code Assist on their own turf? You get Kimi Code CLI, the terminal-based coding agent from Moonshot AI that has racked up 3,592 GitHub stars in less than two months since its debut. And unlike its Western competitors, Kimi Code CLI ships as a single binary that installs in seconds, supports video input out of the box, and connects to the most powerful open-weight model on the coding leaderboards: Kimi K3, with 2.8 trillion parameters.
What Kimi Code CLI Does
Kimi Code CLI is an AI coding agent that runs entirely in your terminal. It reads and edits code, executes shell commands, searches files, fetches web pages, and decides the next step based on the feedback it receives. Think Claude Code or OpenAI Codex CLI, but built from the ground up by the team behind the Kimi K3 model, which currently tops the Arena coding leaderboards. The agent works out of the box with Moonshot AI's own Kimi models and can be configured to use other compatible providers, giving developers flexibility in choosing their underlying model without switching tools.
The project is released under the MIT license and is written in TypeScript, a notable shift from the Python-based original kimi-cli (which is being gradually wound down in favor of this next-generation version). The migration path is smooth: installing Kimi Code CLI automatically migrates your configuration and sessions from the older kimi-cli, so existing users don’t lose their setup.
Key Features That Set It Apart
Kimi Code CLI packs several features that distinguish it from the crowded field of AI coding agents. First is the single-binary distribution model. Installation is a one-liner: curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash on macOS or Linux, brew install kimi-code for Homebrew users, or a PowerShell one-liner on Windows. No Node.js setup, no PATH gymnastics, no global module conflicts. The terminal UI starts in milliseconds, making every session feel instant rather than heavy.
The video input feature is genuinely novel. You can drop a screen recording or demo clip directly into the chat and let the agent watch what is hard to describe in words. Turn a reference clip into a LUT, a long video into a short, a screen recording into working code. No other CLI coding agent offers this capability today.
Kimi Code CLI also introduces AI-native MCP configuration. Instead of hand-editing JSON files to connect Model Context Protocol servers, you use the /mcp-config command conversationally. Add, edit, and authenticate MCP servers without leaving the terminal. The tool also supports ad-hoc MCP configuration via a --mcp-config-file flag for developers who prefer declarative setups.
The subagent system is another standout. Kimi Code CLI lets you dispatch built-in coder, explore, and plan subagents in isolated contexts while keeping the main conversation clean. This means you can run parallel agent tasks for the same project without cross-contamination of context windows.
How It Compares to the Competition
| Feature | Kimi Code CLI | Claude Code | OpenAI Codex CLI |
|---|---|---|---|
| License | MIT (open-weight) | Proprietary | Proprietary / Apache 2.0 |
| Install | Single binary, seconds | npm package | npm package |
| Video Input | Yes, native | No | No |
| MCP Support | AI-native config | JSON config | Limited |
| Subagents | Built-in coder/explore/plan | Custom hooks | No |
| ACP Support | Yes, out of box | Limited | No |
| Startup Time | Milliseconds | 1-2 seconds | 1-3 seconds |
Kimi Code CLI also supports lifecycle hooks that let developers run local commands at key points to gate risky tool calls, audit decisions, trigger desktop notifications, or connect to custom automation. For IDE users, the Agent Client Protocol (ACP) support means you can drive a Kimi Code CLI session straight from Zed, JetBrains, or any ACP-compatible editor using the kimi acp subcommand.
Installation and Quick Start
Getting started with Kimi Code CLI takes less than a minute. On macOS or Linux, run:
curl -fsSL https://code.kimi.com/kimi-code/install.sh | bashOr with Homebrew:
brew install kimi-codeThen navigate to your project and start the agent:
cd your-projectkimiOn first launch, run /login and authenticate via Kimi Code OAuth or a Moonshot AI Open Platform API key. After that, you can start tasks like: "Take a look at this project and explain its main directories." The agent will read the codebase, understand its structure, and provide a comprehensive answer.
For developers who want to contribute, the project uses Node.js 24.15.0+ and pnpm, with a standard pnpm dev:cli development workflow. The test suite, TypeScript checks, and linting are all runnable via pnpm scripts.
Who This Is For
Kimi Code CLI is best suited for developers who want an open-weight coding agent that works instantly, supports advanced features like video input and subagents, and connects to the top-ranked K3 model. It is particularly compelling for developers in the open-source ecosystem who value the MIT license and the ability to configure alternative model providers. Teams working with ACP-compatible editors (Zed, JetBrains) will find the seamless integration a significant productivity boost. If you have been using the older kimi-cli, the automatic migration path makes upgrading painless. For developers who need a coding agent that starts fast, supports parallel agent workflows, and can actually watch video demonstrations, Kimi Code CLI is currently the only option that checks all those boxes.