How to Run the Kimi-Code Application: Installation, CLI Commands, and TUI Walkthrough

Run the kimi binary after installing via the official script, authenticate with /login on first launch, then use the TUI for interactive sessions or CLI flags for one-off tasks.

The kimi-code application from MoonshotAI/kimi-code is a terminal-based AI coding assistant that ships as a self-contained binary. This guide explains how to install, authenticate, and run the application using both its interactive TUI and command-line interface, with references to the actual source implementation in apps/kimi-code.

Installation: Get the Kimi-Code Binary

Kimi-code distributes pre-built binaries through install scripts. Choose your platform:

macOS and Linux

curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash

This script downloads the appropriate binary and places it on your PATH【source 5†https://github.com/MoonshotAI/kimi-code/blob/main/README.md#L14-L20】.

Windows (PowerShell)

irm https://code.kimi.com/kimi-code/install.ps1 | iex

Verify the installation:

kimi --version

This prints the current version and confirms the binary is executable【source 6†https://github.com/MoonshotAI/kimi-code/blob/main/README.md#L32-L34】.

Running Kimi-Code: Two Modes

The application supports interactive TUI sessions and one-off CLI commands. The entry point is apps/kimi-code/src/main.ts, which parses arguments and routes to the appropriate handler.

Interactive TUI Mode (Default)

Navigate to your project directory and run:

cd /path/to/your/project
kimi

This launches the TUI (Terminal User Interface) built on pi-tui【source 1†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/tui/tui-state.ts】. The main loop in src/main.ts initializes the runtime environment, resolves the data directory (~/.kimi-code), loads bundled binaries, and hands control to src/tui/tui-state.ts【source 3†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/main.ts】.

First-Time Authentication

On initial launch, run this command inside the TUI:

/login

Choose OAuth or API-key authentication. The session token persists to the data directory for subsequent runs【source 8†https://github.com/MoonshotAI/kimi-code/blob/main/README.md#L47-L51】.

One-Off CLI Commands

Skip the TUI for quick tasks using run-prompt or sub-agents:

Execute a single prompt:

kimi run-prompt "Explain the main directories of this repo"

This command is implemented in src/cli/run-prompt.ts and processes the request without loading the full UI【source 9†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/cli/run-prompt.ts】.

Use a specialized sub-agent:

kimi coder "Add a unit test for my new function"

Sub-agents (coder, explore, plan) are dispatched from the TUI's command palette and execute focused tasks【source 10†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/tui/utils/foreground-task.ts】.

Runtime Architecture: What Happens When You Run

Understanding the three-layer architecture helps troubleshoot issues when running kim-code:

Layer	Purpose	Key Source File
Bootstrap	Detects platform, resolves $KIMI_CODE_HOME, prepares ~/.kimi-code data and binary directories	src/utils/paths.ts【source 2†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/utils/paths.ts】
Core CLI	Parses --agent/--agent-file flags, routes commands (run-shell, run-prompt, etc.)	src/main.ts【source 3†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/main.ts】
TUI / Runtime	Manages interactive UI, event handling, plugin execution, and feedback loops	src/tui/tui-state.ts【source 4†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/tui/tui-state.ts】

The native asset manager (src/native/native-assets.ts) locates bundled executables like fd and rg in dist-native/bin, enabling tool-augmented responses without external dependencies【source 1†https://github.com/MoonshotAI/kimi-code/blob/main/apps/kimi-code/src/tui/tui-state.ts】.

Advanced: Running from Source

For development or customization, build and run from the repository:

# Clone and install dependencies

git clone https://github.com/MoonshotAI/kimi-code.git
cd kimi-code
npm install

# Build the binary

npm run build

# Run the development build

node apps/kimi-code/dist/main.js

The build process compiles TypeScript and bundles native assets into the dist-native/bin directory.

Platform Support

Kimi-code runs on:

• macOS (Intel and Apple Silicon)
• Linux (x86_64, ARM64)
• Windows (via Git Bash or native PowerShell)

The bootstrap layer in src/utils/paths.ts handles platform-specific path resolution and directory creation automatically.

Summary

• Install with the official curl/PowerShell scripts, then verify with kimi --version
• Authenticate once using /login inside the TUI; credentials persist to ~/.kimi-code
• Run interactively with bare kimi command in any project directory
• Execute one-off tasks with kimi run-prompt "<query>" or kimi coder "<task>"
• Architecture: Bootstrap (paths.ts) → CLI routing (main.ts) → TUI runtime (tui-state.ts)

Frequently Asked Questions

Do I need to authenticate every time I run kimi-code?

No. Run /login once inside the TUI on first launch. The session token persists to ~/.kimi-code as handled by the authentication flow in src/tui/tui-state.ts. Subsequent launches use the cached credentials automatically.

Can I use kimi-code without the interactive TUI?

Yes. Use kimi run-prompt "<your question>" for single queries, or invoke sub-agents directly like kimi coder "<task description>". These commands execute via src/cli/run-prompt.ts without initializing the full terminal interface.

Where does kimi-code store its data and bundled binaries?

Runtime data lives in ~/.kimi-code (or $KIMI_CODE_HOME if set). Bundled native binaries like fd and rg are managed by src/native/native-assets.ts and extracted to dist-native/bin within the application directory. Platform detection and path resolution occur in src/utils/paths.ts.

What sub-agents are available when running kimi-code?

The main sub-agents include coder (code generation/editing), explore (repository exploration), and plan (task planning). These are dispatched through the foreground task system in src/tui/utils/foreground-task.ts and can be invoked directly from the CLI or selected in the TUI command palette.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}