moonshotai/kimi-code is an AI coding agent CLI that runs in the terminal. It does more than read and edit code: it runs shell commands, searches files, fetches web pages, and decides the next step on its own based on the results. Moonshot AI's Kimi models are available out of the box, and other compatible providers can be configured. The latest public release is @moonshot-ai/kimi-code@2.2.0, the implementation language is TypeScript, and the license is MIT.
It supports many extension points: a dedicated TUI (an interface that draws the terminal screen directly), video input, MCP (Model Context Protocol, a convention that lets agents connect to external tools) configuration, plugins, sub-agents, and lifecycle hooks. It can also be used inside editors and IDEs such as Zed and JetBrains through ACP, the convention for calling agents within an editor. This repository was created on 2026-05-22 and has 7939 stars.
That said, the public README contains no performance figures such as task success rate or response latency, and no comparison with other coding agents. On the evaluation scores, the Momentum item is 3.3 out of 10, and no values are given for the other three items. There is no basis for claiming a performance advantage. It looks like a reasonable scope to check first for people who already use Kimi models or work in Zed or JetBrains environments. Any further judgment becomes possible only after more material accumulates.