Moonshot AI 發布 Kimi Code CLI：以 TypeScript 打造的終端機 AI 編碼代理，專為次世代代理設計

Moonshot AI has released Kimi Code CLI, an open-source coding agent that runs in the terminal. The tool reads and edits code, runs shell commands, searches files, and fetches web pages. It then chooses its next step based on the feedback it receives. The project is MIT-licensed and lives on GitHub.. Kimi Code CLI is the successor to the older kimi-cli. The new agent is written in TypeScript and distributed via npm. It works out of the box with Moonshot AI’s Kimi models. It can also be configured to use other compatible providers. What is Kimi Code CLI Kimi Code CLI is an AI agent for software development and terminal operations. It can implement new features, fix bugs, and complete refactors. It can also explore an unfamiliar codebase and answer architecture questions. Batch file processing, builds, and chained test runs are supported too. The execution model is feedback-driven. The agent plans steps, modifies code, runs tests, and reports its actions. Read-only operations run automatically by default. For file edits or shell commands, the agent asks for confirmation first. This approval flow keeps risky actions under developer control. The CLI itself is free and MIT-licensed. Model access requires Kimi Code OAuth or a Moonshot AI Open Platform API key. https://github.com/MoonshotAI/kimi-code Key Features Moonshot lists several features aimed at long, focused agent sessions: Single-binary distribution. One command installs it, with no Node.js setup required. Fast startup. Moonshot says the TUI is ready in milliseconds. Purpose-built TUI. The interface is tuned for extended agent sessions. Video input. Drop a screen recording or demo clip into the chat. AI-native MCP configuration. Add and authenticate Model Context Protocol servers via /mcp-config. Subagents for parallel work. Dispatch built-in coder, explore, and plan subagents in isolated contexts. Lifecycle hooks. Run local commands to gate tool calls, audit decisions, or trigger notifications. Installation and First Run Two installation paths exist. The official script needs no pre-installed Node.js. On macOS or Linux, run the install script: Copy CodeCopiedUse a different Browsercurl -fsSL https://code.kimi.com/kimi-code/install.sh | bash On Windows, use PowerShell: Copy CodeCopiedUse a different Browserirm https://code.kimi.com/kimi-code/install.ps1 | iex The global npm install requires Node.js 24.15.0 or later: Copy CodeCopiedUse a different Browsernpm install -g @moonshot-ai/kimi-code Verify the binary, then open a project and start the interactive UI: Copy CodeCopiedUse a different Browserkimi --version cd your-project kimi On first launch, type /login inside the UI. You can choose Kimi Code OAuth or a Moonshot AI Open Platform API key. To run one instruction without the UI, use kimi -p "your task". To resume the previous session, add -C. Use Cases Understanding a project: Ask for an architecture overview and a module dependency diagram. Implementing a feature: Describe the signature, options, and acceptance criteria up front. Fixing a bug: Give the symptom, reproduction steps, and expected behavior together. Writing tests and refactoring: Extract repeated patterns, then run tests to confirm behavior. One-off automation: Analyze logs and output call counts with p50 and p99 latencies. Scheduled tasks: Ask the agent to set reminders or recurring checks via cron. Plan mode is available through Shift-Tab or kimi --plan. It outputs a research plan before touching files. For safe batch work, --yolo or /yolo skips approval prompts. The /fork command creates an experimental branch you can abandon. The /compact command compresses context to free up tokens. For large investigations, the main agent can dispatch subagents in parallel. How Kimi Code CLI Compares Kimi Code CLI joins several established terminal coding agents. The table below compares it with three of them. Competitor details reflect mid-2026 and can change quickly. AttributeKimi Code CLIClaude CodeOpenAI Codex CLIGemini CLIDeveloperMoonshot AIAnthropicOpenAIGoogleBacking modelKimi modelsClaude modelsGPT-5.3-CodexGemini 2.5 ProLanguage / runtimeTypeScriptNode.jsRustTypeScriptInstallScript or npm (Node.js ≥ 24.15.0)Native installer or npmnpm / nativenpm single binaryMCP supportYes (/mcp-config)YesYesYesSubagentsYes (coder, explore, plan)YesYesNo (sequential)Plan modeYes (Shift-Tab)YesYesYesIDE integrationACP (Zed, JetBrains)VS Code, JetBrainsVS Code, IDEsVS Code (Code Assist)LicenseMITProprietaryOpen sourceApache 2.0 All four agents support the Model Context Protocol. They differ on backing model, language, license, and orchestration. Kimi Code CLI and Codex CLI both ship native subagents. Gemini CLI runs tasks sequentially without subagent support. Key Takeaways Kimi Code CLI is an MIT-licensed terminal coding agent from Moonshot AI. It is written in TypeScript and installs via script or npm. Built-in coder, explore, and plan subagents run in isolated contexts. MCP servers are configured conversationally through /mcp-config, not raw JSON. It succeeds kimi-cli and migrates existing configuration and sessions. Marktechpost’s Visual Explainer #mtp-kimi-slider *{box-sizing:border-box!important;margin:0;padding:0} #mtp-kimi-slider{ --mtp-bg:#ffffff;--mtp-ink:#000000;--mtp-ink-2:#333333;--mtp-ink-3:#666666; --mtp-line:#000000;--mtp-soft:#e6e6e6;--mtp-fog:#f4f4f4; font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif!important; max-width:760px;margin:24px auto!important;background:var(--mtp-bg)!important;color:var(--mtp-ink)!important; border:2px solid var(--mtp-line)!important;border-radius:0!important;overflow:hidden!important; -webkit-font-smoothing:antialiased;line-height:1.5 } /* wpautop suppression */ #mtp-kimi-slider hr,#mtp-kimi-slider p:empty,#mtp-kimi-slider del,#mtp-kimi-slider s{display:none!important} /* Header */ #mtp-kimi-slider .mtp-head{ display:flex;align-items:center;justify-content:space-between; background:var(--mtp-ink)!important;color:#fff!important; padding:14px 18px!important;border-bottom:2px solid var(--mtp-line)!important } #mtp-kimi-slider .mtp-kicker{ font-size:12px;font-weight:800;letter-spacing:.16em;text-transform:uppercase;color:#fff!important } #mtp-kimi-slider .mtp-count{ font-size:12px;font-weight:700;letter-spacing:.08em;color:#fff!important; border:1px solid #fff!important;padding:3px 9px!important;border-radius:0!important } /* Viewport / track */ #mtp-kimi-slider .mtp-view{overflow:hidden!important;background:var(--mtp-bg)!important} #mtp-kimi-slider .mtp-track{display:flex!important;transition:transform .35s ease} #mtp-kimi-slider .mtp-slide{ flex:0 0 100%!important;min-width:100%!important;padding:30px 30px 26px!important;background:var(--mtp-bg)!important } #mtp-kimi-slider .mtp-num{ font-size:12px;font-weight:800;letter-spacing:.14em;text-transform:uppercase;color:var(--mtp-ink-3)!important;margin-bottom:8px!important } #mtp-kimi-slider h2.mtp-title{ font-size:26px;line-height:1.18;font-weight:800;color:var(--mtp-ink)!important;margin:0 0 14px!important;letter-spacing:-.01em } #mtp-kimi-slider .mtp-sub{font-size:16px;color:var(--mtp-ink-2)!important;margin:0 0 12px!important} /* Bullets */ #mtp-kimi-slider ul.mtp-list{list-style:none!important;margin:6px 0 0!important;padding:0!important} #mtp-kimi-slider ul.mtp-list li{ position:relative;padding:7px 0 7px 22px!important;font-size:15.5px;color:var(--mtp-ink-2)!important; border-bottom:1px solid var(--mtp-soft)!important } #mtp-kimi-slider ul.mtp-list li:last-child{border-bottom:none!important} #mtp-kimi-slider ul.mtp-list li:before{ content:"";position:absolute;left:0;top:14px;width:9px;height:9px;background:var(--mtp-ink)!important } #mtp-kimi-slider code{ font-family:"SF Mono",ui-monospace,Menlo,Consolas,monospace!important; background:var(--mtp-fog)!important;color:var(--mtp-ink)!important;border:1px solid var(--mtp-soft)!important; padding:1px 5px!important;font-size:13.5px;border-radius:0!important } #mtp-kimi