Kimi Code CLI

license

MIT(OSI)

source

public

core features withheld

no — the tree at 4a82a07 carries one root MIT LICENSE and no ee/, enterprise/, commercial/ or proprietary/ path, and the workspace packages the CLI bundles declare MIT or fall under that root file. Remote Control, which needs a paid Kimi membership, relays through a hosted Kimi service

Kimi Code CLI is MIT licensed, and its public repository holds the CLI and every package it bundles, so it builds and runs from published source. A Kimi Code account buys model access, and the CLI also runs against Anthropic, OpenAI-compatible and Gemini providers. Remote Control is limited to paid members because it relays through Moonshot AI's hosted service, but its client is published and the agent still runs on the user's machine.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/configuration/providers.md recorded 2026-10-19

  The providers page says the CLI "supports connecting to multiple LLM platforms simultaneously: one-click login via the Kimi Code managed service, connecting Claude with an Anthropic API key, or connecting third-party inference services via the OpenAI-compatible protocol".

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/guides/remote-control.md recorded 2026-10-19

  The Remote Control guide says "Remote Control requires a paid membership and is not available to free users" and "Tasks always run on your machine — the web page is just a remote window."

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/LICENSE recorded 2026-10-19

  The root LICENSE is the MIT License text, "Copyright (c) 2026 Moonshot AI".

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/pnpm-workspace.yaml recorded 2026-10-19

  The workspace file lists packages/*, apps/*, apps/vis/server, apps/vis/web and docs as the repository's own packages.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/README.md recorded 2026-10-19

  The README documents cloning MoonshotAI/kimi-code and running "pnpm build" to build all packages, and closes "Released under the [MIT License](LICENSE)."; no paid edition or license key is mentioned.

• https://registry.npmjs.org/@moonshot-ai/kimi-code/latest recorded 2026-10-19

  The npm record for @moonshot-ai/kimi-code 2.2.0 declares "license": "MIT", a repository of MoonshotAI/kimi-code under apps/kimi-code, and the kimi command.

Adoption is the trailing month of npm downloads of @moonshot-ai/kimi-code. The recommended install script downloads a standalone binary from Moonshot AI's own server, and those installs and Homebrew's are not counted, so the figure is a lower bound on use. Downloads of the archived Python kimi-cli are left out because they install the predecessor, not this tool.

• https://api.npmjs.org/downloads/point/last-month/@moonshot-ai/kimi-code recorded 2026-10-19

  169,243 downloads of @moonshot-ai/kimi-code between 2026-09-17 and 2026-10-16

• https://code.kimi.com/kimi-code/install.sh recorded 2026-10-19

  The install script sets KIMI_DOWNLOAD_BASE="https://code.kimi.com/kimi-code" and fetches the binary from its binaries path, not from npm.

• https://pypi.org/pypi/kimi-cli/json recorded 2026-10-19

  The PyPI record for kimi-cli 1.52.0 carries the summary "[Archived] Legacy Python Kimi CLI, no longer maintained. Use Kimi Code CLI instead" and the classifier "Development Status :: 7 - Inactive".

• https://pypistats.org/api/packages/kimi-cli/recent recorded 2026-10-19

  50,722 downloads of kimi-cli in the last month

Kimi Code CLI covers the core of a terminal coding agent and adds built-in sub-agents, an experimental mode that runs parallel workers in separate git worktrees with independent review, lifecycle hooks and editor integration over ACP. It sits a step below OpenHands, which runs agents in a sandboxed runtime and publishes benchmark results, while Kimi Code CLI runs with the user's own permissions and has no benchmark of its own.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/configuration/providers.md recorded 2026-10-19

  The provider table lists kimi, anthropic, openai, openai_responses, google-genai and vertexai types.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/customization/agents.md recorded 2026-10-19

  The agents page says "Kimi Code CLI includes three built-in sub-agents": coder, explore and plan, each working in its own isolated context.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/guides/getting-started.md recorded 2026-10-19

  The getting-started guide opens the interactive UI with kimi and runs a single instruction with kimi -p: "To run a single instruction without entering the interactive UI, use `-p`".

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/guides/interaction.md recorded 2026-10-19

  The interaction guide describes Plan mode, Ask When Needed mode and the fully unattended Never Ask mode for tool approvals.

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/guides/sessions.md recorded 2026-10-19

  The sessions guide says Kimi Code CLI persists every conversation as a session and covers how to "resume sessions, manage context, and export or fork sessions".

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/guides/tower.md recorded 2026-10-19

  "Tower is an experimental multi-agent collaboration mode"; "Every mission has its own branch and Git worktree", and "Worktrees isolate writes, but they are not a security sandbox."

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/docs/en/reference/tools.md recorded 2026-10-19

  The built-in tools reference lists Read, Write, Edit, Grep, Glob and ReadMediaFile, shell tools, and AgentSwarm, which "launches subagents from a shared prompt_template and an items array".

• https://raw.githubusercontent.com/MoonshotAI/kimi-code/main/README.md recorded 2026-10-19

  The README lists "Subagents for focused, parallel work", "Lifecycle hooks", "AI-native MCP configuration", "Video input" and "Editor & IDE integration (ACP)", and says the agent "can also be configured to use other compatible providers"; no benchmark figure.