Skip to content

README.md

Latest commit

 

History

History
284 lines (208 loc) · 8.49 KB

README.md

File metadata and controls

284 lines (208 loc) · 8.49 KB
Raw

Kimi Code CLI Introduction

What is Kimi Code CLI?

Kimi Code CLI is Moonshot AI's AI coding agent that runs in your terminal. It can read and edit code, run shell commands, search files, fetch web pages, and decide the next step based on feedback from tools and the user. It works out of the box with Kimi Code / Moonshot AI models and can also be configured for other compatible providers.

Core Philosophy

Terminal-first agent workflow

Kimi Code is designed for developers who want to stay in the terminal:

  • Start an interactive TUI directly in a project directory
  • Ask natural-language questions about a repository
  • Let the agent plan, edit, run tests, and iterate in the same session

Controlled autonomy

Kimi Code can do more than answer questions, but keeps permission modes explicit:

  • Read-only exploration runs smoothly by default
  • File edits and shell commands can require approval
  • YOLO and Auto modes are available when you deliberately want less friction

Extensible agent platform

Kimi Code includes capabilities commonly needed for agentic development:

  • MCP configuration through natural-language commands
  • Skills and subagents for focused tasks
  • Lifecycle hooks for local automation
  • ACP editor integration for Zed, JetBrains, and other Agent Client Protocol clients

Core Features

Code understanding and editing

  • Analyze project structure and dependencies
  • Read, search, and modify files
  • Implement features, fix bugs, refactor code, and generate tests

Shell and workflow automation

  • Run build, test, lint, and project scripts
  • Chain multiple steps while observing tool feedback
  • Support non-interactive prompt mode for scripts and CI-style tasks

Multimodal and editor integration

  • Paste screenshots or videos into the TUI when the selected model supports multimodal input
  • Use kimi acp from ACP-compatible editors and IDEs
  • Open the browser-based UI via the local Kimi server

Quick Start

Installation

Recommended install script:

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

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

Homebrew:

brew install kimi-code

npm / pnpm installation:

npm install -g @moonshot-ai/kimi-code
# or
pnpm add -g @moonshot-ai/kimi-code

Verify the installation:

kimi --version

First launch

cd your-project
kimi

On first launch, run /login in the TUI and choose either Kimi Code OAuth or a Moonshot AI platform API key. You can also log in from the shell without entering the TUI:

kimi login

Common CLI Commands

# Start an interactive session in the current directory
kimi

# Continue the most recent session in the current directory
kimi --continue
kimi -C

# Choose a previous session, or resume a known session ID
kimi --session
kimi --session <session-id>

# Run one prompt non-interactively
kimi --prompt "Summarize the current repository status"
kimi -p "Explain the latest diff"

# Emit JSONL-style streaming output for automation
kimi -p "List changed files" --output-format stream-json

# Use a specific model for this launch
kimi --model kimi-code/kimi-for-coding
kimi -m kimi-code/kimi-for-coding -p "Explain this project"

# Start in Plan mode, focusing on read-only exploration before edits
kimi --plan

# Add an extra workspace directory for the session
kimi --add-dir ../shared-lib

# Load skills from custom directories for this launch
kimi --skills-dir /path/to/team-skills --skills-dir ./local-skills

Permission Modes: Manual, YOLO, Auto, and Plan

Kimi Code's normal interactive flow asks before side-effecting operations such as file writes and shell commands. You can switch modes when you need a different balance between safety and speed.

YOLO mode

YOLO mode skips approval prompts for regular tool calls, including file writes and shell command execution. Use it only in trusted repositories and for tasks where you are comfortable reviewing the result afterward.

# Start a session with YOLO mode enabled
kimi --yolo
kimi -y

# Hidden aliases documented by Kimi Code
kimi --yes
kimi --auto-approve

Inside the TUI:

/yolo
/yolo on
/yolo off
/yes

Plan-mode exit approval is not bypassed by YOLO mode.

Auto mode

Auto mode lets the agent handle approvals automatically and avoids asking the user clarifying questions. It is useful for unattended work where you want more autonomy without using the full YOLO toggle.

kimi --auto
kimi --continue --auto

Inside the TUI:

/auto
/auto on
/auto off

Plan mode

Plan mode asks the agent to explore and produce a plan before making file changes. It is useful for larger or higher-risk tasks.

kimi --plan

Inside the TUI:

/plan
/plan clear

Keyboard shortcut: Shift-Tab toggles Plan mode.

Useful Slash Commands

Command Purpose
/help Show built-in help, shortcuts, and available commands
/login / /logout Sign in or clear credentials
/provider Manage configured model providers
/model Switch the model used in the current session
/permission Select a permission mode
/settings or /config Open settings inside the TUI
/new or /clear Start a fresh session
/sessions or /resume Browse and restore session history
/compact Compress the current context to free tokens
/undo Remove recent prompts from active context
/reload Reload configuration without restarting the CLI
/init Analyze the codebase and generate AGENTS.md
/add-dir Add an extra workspace directory
/mcp List MCP server status
/mcp-config Configure MCP servers and MCP OAuth login
/plugins Open the plugin manager
/editor Configure the external editor opened by Ctrl-G
/theme Change the terminal UI theme
/exit, /quit, /q Exit Kimi Code CLI

Keyboard Shortcuts

Shortcut Function
Enter Submit the current input
Shift-Enter / Ctrl-J Insert a newline
↑ / ↓ Browse input history
Esc / Ctrl-C Interrupt streaming output or close a popup
Ctrl-D Exit when the input box is empty
Shift-Tab Toggle Plan mode
Ctrl-S Inject a message into a running turn
Ctrl-O Collapse or expand tool output
Ctrl-G Edit the current input in an external editor

Configuration Files

Kimi Code stores local data under ~/.kimi-code/ by default. Set KIMI_CODE_HOME to move this directory.

Main files:

  • ~/.kimi-code/config.toml - providers, models, default model, permission mode, hooks, and agent/runtime settings
  • ~/.kimi-code/tui.toml - terminal UI preferences such as theme, editor, notifications, and update behavior
  • ~/.kimi-code/mcp.json - MCP server declarations
  • .kimi-code/local.toml - project-local remembered settings such as extra workspace directories

Useful maintenance commands:

kimi doctor
kimi doctor config
kimi doctor tui
kimi upgrade

Server, Web UI, and ACP

# Start or reuse the local background server
kimi server run

# Open the browser UI
kimi web

# Run the server in the foreground
kimi server run --foreground

# Entry point used by ACP-compatible editors
kimi acp

For Zed or JetBrains ACP integration, configure the editor to run kimi acp over stdio after logging in once from the CLI.

Best Practices

  1. Start in Plan mode for large changes - Review the plan before letting the agent edit files.
  2. Use YOLO only in trusted directories - It skips approval for edits and shell commands.
  3. Keep prompts concrete - Mention files, expected behavior, tests to run, and constraints.
  4. Review diffs and run tests - Treat agent output like a teammate's patch.
  5. Use /mcp-config for integrations - Prefer the built-in MCP flow over hand-editing JSON when possible.
  6. Use -p for automation - Non-interactive prompt mode is better for scripts than driving the TUI.

Related Resources

License

Kimi Code CLI is released under the MIT License. Usage of Kimi Code, Moonshot AI, or third-party model providers is subject to the corresponding provider terms.