Unofficial community guide for Kimi Code CLI. Not affiliated with Moonshot AI. All trademarks belong to their owners.
Kimi Code CLI is the terminal version of Kimi Code, Moonshot AI's coding agent. Kimi Code ships in three forms, desktop, CLI and IDE plugins, under one subscription, and the CLI is the one you run inside a shell next to your repository. Moonshot describes the agent as able to write and edit code, understand a codebase by searching and reading it, run commands such as tests and builds and verify the results, and look things up on the web. This guide is written from the Kimi Code product page, the Kimi API Platform integration guide for the CLI, and the launch coverage that described it as a terminal AI coding agent built in TypeScript.
Do not need an agent session at all, just a working site from a prompt? Try Begin.sh - turn a prompt or a URL to clone into a static site or Expo app and download the zip. More on when that fits below.
The product page's own demo shows the shape of a session: a sidebar of projects (a landing page, an app, an API server), each with a list of past sessions such as hero copy polish, fix session list scroll jitter, or auth middleware refactor, and a chat pane where the agent introduces itself and lists what it can do. That is the same model whether you use the desktop app, the CLI or an IDE plugin; the CLI simply puts it in the terminal, which is where it fits into scripts, tmux and remote boxes. The launch write-up on Marktechpost dated 6 June 2026 describes it as a terminal AI coding agent written in TypeScript and aimed at next-generation agent workflows.
The model behind it is Kimi K3, which the product page says supports up to 1M context tokens and is available on the higher membership plans. The Kimi API Platform docs also cover Kimi K2.7 (a code model) and K2.6, so you can expect the CLI to be usable with more than one Kimi model when you bring your own API key. The platform docs list the capabilities those models expose: thinking models, reasoning effort, streaming, JSON mode, partial mode, vision input, context caching, dynamic tool loading, tool calls, built-in web search and a batch API.
- Decide how you will pay. Route A is a Kimi membership from the pricing page; Kimi Code is included in the plans shown on the Kimi Code page, and K3 is only on some tiers. Route B is a Kimi API Platform key, following Use Kimi API Platform with Kimi Code CLI.
- Read the Kimi Code docs for the install command and first run; the exact command is in the docs rather than on the marketing page, so take it from there.
- If you chose route B, create the key in the User Center and check Models and Pricing so you know what a long session costs.
- Run the CLI inside a git repository. Everything the agent does (edits, commands) is easier to review as a diff.
- Start with a small, verifiable task: a failing test, a lint error, a copy change. Let it run the test command itself and check that it reports the result honestly.
Two price lists apply. Membership tiers are on the Kimi pricing page, and the product page notes that buying a plan also unlocks other Kimi membership benefits. API usage is on the platform pricing page. Neither number is reproduced here because they change; check the pages.
- K3 is gated by plan tier. The product page says K3 is available on certain plans and above, so a cheaper membership may put you on an older model. Confirm which model the CLI is actually using before judging its output.
- 1M context is the ceiling, not the default cost. Long sessions that read many files use context quickly; start a fresh session for unrelated tasks.
- Context caching is listed in the platform docs. If you go the API-key route, it is the feature that makes repeated long-context calls affordable.
- The platform also has guides for using Kimi models inside Claude Code, Codex, OpenCode, OpenClaw and Hermes Agent. If your team already standardised on one of those, you can keep the tool and swap the model instead of adopting a new CLI.
- MoonPalace, the API debugging tool in the platform docs, is worth pointing the CLI's traffic through when a session behaves oddly.
- The Kimi Code page is at kimi.com; the API platform is at platform.kimi.ai (the open platform link goes to platform.kimi.com). Keys and subscriptions are separate systems.
| Kimi Code CLI | Kimi Code desktop and IDE plugins | Begin.sh | |
|---|---|---|---|
| Where it runs | Your terminal | Desktop app, IDE | Browser |
| What you get | An agent editing your repo and running commands | Same agent with a GUI | A downloadable zip of a static site or Expo app |
| Input | Conversation plus your codebase | Conversation plus your codebase | A prompt, or a URL to clone |
| Model | Kimi K3 on eligible plans, or your API key | Same | Not exposed |
| Hosting, backend, auth | Whatever you build | Whatever you build | None included |
Is Kimi Code CLI free? It is part of Kimi Code, which the product page ties to membership plans. Check the pricing page; an API key is the alternative.
Which model does it use? Kimi K3 with up to 1M context on eligible plans. With an API key, the platform model list applies.
Does it work with other coding agents? Not the CLI itself, but the platform has guides for using Kimi models in Claude Code, Codex, OpenCode, OpenClaw and Hermes Agent.
Where is the install command? In the Kimi Code docs at kimi.com/code/docs/en and in the platform integration guide; it is not on the landing page.
Is it open source? The launch coverage says it is built in TypeScript; the pages covered here do not state a licence, so check the docs.
Kimi Code CLI is for working inside an existing codebase with an agent. If the task is the opposite, you have no codebase yet and want a marketing site or a small Expo app you can look at in five minutes, Begin.sh takes a prompt or a URL to clone and gives you a zip of a working static site or Expo app. It ships no hosting, backend or auth, which is exactly right for a prototype you will then hand to an agent like Kimi Code to grow.
Try Begin.sh - prompt to static site or Expo app, download the zip
Last reviewed: 2026-09-22