Moonshot AI's Kimi CLI Evolves into Kimi Code CLI: A Next-Gen Terminal AI Agent

What Changed

Moonshot AI is evolving its Kimi CLI, an AI agent designed for terminal-based software development tasks, into the next-generation Kimi Code CLI. The new Kimi Code CLI is positioned as a more advanced terminal AI agent from the same development team. Users installing Kimi Code CLI will experience an automatic migration of their existing configurations and sessions from Kimi CLI. While the Kimi CLI project will be gradually phased out, its documentation and current installations will remain accessible.

Kimi CLI, and by extension Kimi Code CLI, functions as an AI agent within the terminal environment. Its core capabilities include assisting with software development tasks and general terminal operations. This encompasses reading and editing code, executing shell commands, performing web searches and fetching web pages, and autonomously planning and adapting actions during execution.

Technical Details

Kimi Code CLI builds upon the foundation of Kimi CLI, offering several key features designed to streamline developer workflows:

Shell Command Mode: The agent integrates a shell command mode, accessible by pressing Ctrl-X. This allows users to execute shell commands directly within Kimi CLI without needing to exit the environment. It's noted that built-in shell commands like cd are not yet supported.

VS Code Extension: Kimi CLI offers integration with Visual Studio Code through the Kimi Code VS Code Extension. This facilitates a unified experience for developers using VS Code as their primary IDE.

IDE Integration via ACP: The agent supports the Agent Client Protocol (ACP) natively. This enables integration with any ACP-compatible editor or IDE. To utilize this feature, users must run Kimi CLI in the terminal, complete a /login command, and then configure their ACP client to start Kimi CLI as an ACP agent server using the kimi acp command. Examples provided include integration with Zed and JetBrains IDEs, requiring specific JSON configurations in their respective settings files.

Zsh Integration: For users of the Zsh shell, Kimi CLI can be integrated via the zsh-kimi-cli plugin. Installation involves cloning the plugin repository and adding kimi-cli to the Zsh plugin list in the ~/.zshrc file. This integration empowers the shell experience with AI agent capabilities, allowing users to switch to agent mode by pressing Ctrl-X.

MCP Support: Kimi CLI incorporates support for Model Context Protocol (MCP) tools. This includes a kimi mcp sub-command group for managing MCP servers. Users can add streamable HTTP servers (with or without OAuth authorization) and stdio servers, list existing servers, remove them, and authorize access. Additionally, ad-hoc MCP server configuration is supported via the --mcp-config-file CLI option, allowing connection to specified MCP servers using a well-known MCP config file format.

For development, the project outlines a standard Git workflow. After cloning the repository and running make prepare, developers can use uv run kimi to execute the CLI. The Makefile includes targets for formatting code (make format), running linting and type checking (make check), executing tests (make test, make test-kimi-cli, make test-kosong, make test-pykaos), building the web UI (make build-web), and building Python packages (make build) or standalone binaries (make build-bin). The make build and make build-bin commands automatically incorporate the web UI build.

Developer Implications

The transition to Kimi Code CLI and its comprehensive feature set have several implications for developers. The automatic migration of configurations and sessions aims to ensure a smooth transition for existing Kimi CLI users, minimizing setup overhead. The deep integration with popular development environments like VS Code and ACP-compatible IDEs (Zed, JetBrains) suggests a focus on embedding AI assistance directly into developers' existing workflows, potentially reducing context switching.

The shell command mode and Zsh integration indicate an effort to make the AI agent a more integral part of the command-line experience, extending its utility beyond just coding tasks to general terminal operations. The support for Agent Client Protocol (ACP) and Model Context Protocol (MCP) signifies an open and extensible architecture, allowing developers to connect Kimi CLI with a broader ecosystem of tools and services. This extensibility could enable custom integrations and specialized workflows, catering to diverse development needs.

The development process, outlined with make commands for testing, building, and formatting, suggests a structured approach to maintaining the project. This can be beneficial for contributors looking to extend or customize the agent.

Bottom Line

Moonshot AI's evolution from Kimi CLI to Kimi Code CLI represents a strategic move to provide a more integrated and capable AI agent for software development within the terminal. The agent's ability to handle code, execute shell commands, and integrate with major IDEs and protocols like ACP and MCP positions it as a versatile tool for developers. While the Kimi CLI project is being deprecated, the continuity provided by automatic migration and the enhanced features of Kimi Code CLI aim to offer a robust solution for AI-assisted development workflows. The emphasis on extensibility and integration points to a future where AI agents play a more central role in the developer's daily toolkit.