Kimi Code

Kimi Code is Moonshot AI’s MIT-licensed coding agent for terminal and editor workflows. The current project runs as a single-binary terminal interface under the kimi command and does not require Node.js. It can read and edit files, execute shell commands, search a repository, fetch web pages, and choose its next step from tool results. The canonical MoonshotAI/kimi-code repository supersedes the earlier MoonshotAI/kimi-cli project, which is being wound down with migration support for existing configuration and sessions.

Official shell and PowerShell installers cover Unix-like systems and Windows. On first launch, /login offers Kimi Code OAuth or a Moonshot AI Open Platform API key, and the agent can be configured for other compatible providers. Access charges therefore depend on the selected account or provider. Current allowances and provider terms belong on the live service and billing pages rather than in a durable directory record.

Top features

• Terminal workspace: inspect a repository, edit files, run commands, search code, fetch web content, and review results in a focused TUI.
• Built-in subagents: dispatch coder, explore, and plan workers in isolated contexts for focused or parallel tasks.
• ACP editor integration: run the agent through Zed, JetBrains, or another Agent Client Protocol client with the kimi acp command.
• Plugin system: install skills, MCP servers, and data sources from the marketplace or a GitHub repository, with trust information shown during installation.
• Conversational MCP setup: add, edit, and authenticate Model Context Protocol servers through /mcp-config instead of hand-editing configuration files.
• Video input: attach a screen recording or demo clip when visual behavior is easier to show than describe.
• Hooks and provider configuration: customize lifecycle behavior and connect Kimi models or compatible model services.

Use cases

• Implement a contained feature, run its tests, and revise the patch from terminal output.
• Explore an unfamiliar repository with the explore subagent while keeping the main session focused on implementation.
• Draft a plan, delegate bounded coding work, and review the resulting edits in one terminal session.
• Drive the same agent from Zed, JetBrains, or another ACP-compatible editor.
• Connect approved documentation, data, or development tools through MCP and plugins.
• Use a screen recording to reproduce an interface problem or translate a visual reference into code.