跳到主要內容

Goose 開源 AI Agent 完整指南:Block 打造的本地端自動化工程助手 | Goose Open-Source AI Agent Guide: Block's Local-First Engineering Automation Tool

By Kit 小克 | AI Tool Observer | 2026-04-10

🇹🇼 Goose 開源 AI Agent 完整指南:Block 打造的本地端自動化工程助手

Goose 是 Block(前 Square)開發的開源 AI Agent,能在本地端自動執行程式碼安裝、編輯、測試等工程任務,支援 15 種以上 LLM 供應商,目前已加入 Linux Foundation 的 Agentic AI Foundation。

Goose 是什麼?為什麼開發者需要關注?

如果你還在用 AI 助手幫你「建議」程式碼,Goose 要你重新想像 AI 在開發流程中的角色。Goose 不只是一個 code completion 工具,它是一個真正的 AI Agent——能自己讀寫檔案、跑測試、裝套件、甚至在你的終端機裡自主執行多步驟任務。

Block 的 CTO Dhanji Prasanna 說得很直白:「讓 Goose 開源,是為了創造一個讓發明和成長達到新高度的框架。」而 Block 內部的工程師已經在用 Goose 減少重複性工作,把時間留給更有價值的事。

Goose 的核心功能有哪些?

  • 自主任務執行:不只建議,直接動手——搜尋、導航、寫 code、跑測試、裝依賴,一條龍完成
  • 模型自由選擇:支援 Anthropic、OpenAI、Google、Ollama、OpenRouter、Azure、Bedrock 等 15+ 供應商,不綁定任何一家
  • MCP 原生整合:透過 Anthropic 的 Model Context Protocol 連接 70+ 擴充套件,動態發現新系統和工具
  • 跨平台原生應用:macOS、Linux、Windows 都有桌面版,也有完整 CLI 和 API
  • Rust 打造:用 Rust 寫的核心(佔 58.6%),效能好、跨平台移植性強
  • 本地執行:所有操作在你的機器上跑,資料不離開本地端,隱私和延遲問題一次解決

Goose 跟 Claude Code、Cursor 有什麼不同?

市面上 AI 開發工具百花齊放,Goose 的定位其實很明確:

  • vs Claude Code:Claude Code 是 Anthropic 的終端 AI Agent,綁定 Claude 模型。Goose 是模型無關的(model-agnostic),你可以接任何 LLM
  • vs Cursor:Cursor 是 VS Code fork,走 IDE 整合路線。Goose 是獨立 Agent 框架,有桌面 app、CLI、API 三種介面
  • vs GitHub Copilot:Copilot 主打 code suggestion。Goose 主打「action-oriented」——不是建議你怎麼改,而是直接改好給你看

最大差異?Goose 是完全開源(Apache 2.0),你可以自由修改、商用、部署,沒有授權費。

怎麼安裝 Goose?

CLI 一行指令搞定:

curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash

桌面版直接到 goose-docs.ai 下載 macOS、Linux 或 Windows 安裝檔。裝好後選你的 LLM 供應商、填 API key,就能開始用了。

Goose 加入 Linux Foundation 代表什麼?

2025 年底,Linux Foundation 成立了 Agentic AI Foundation(AAIF),三個創始專案分別是:Anthropic 的 MCP、Block 的 Goose、OpenAI 的 AGENTS.md。白金會員包括 AWS、Google、Microsoft、Anthropic、OpenAI、Cloudflare 等大咖。

這代表 Goose 不再只是 Block 的內部工具,而是 AI Agent 生態系的基礎建設之一。MCP 負責標準化 AI 與工具的連接,Goose 負責提供開源 Agent 框架,AGENTS.md 負責讓 AI Agent 理解各個專案的規範。三者搭配,基本上就是 AI Agent 時代的「Linux + Docker + README」。

實際使用體驗如何?

GitHub 上 40.8k 星、4k fork、127 個 release(最新 v1.30.0,2026 年 4 月 8 日),社群活躍度很高。實際用起來:

  • 設定簡單,選 provider、填 key 就能用
  • MCP 擴充生態系豐富,70+ extensions 可以接各種開發工具
  • Rust 核心確實快,回應感覺比 Python 為主的工具俐落
  • 模型選擇自由度高,想用 Claude 就用 Claude,想跑本地 Ollama 也行

要注意的是,Agent 類工具的效果極度依賴底層模型的能力。用頂級模型(Claude Opus、GPT-5.4)體驗很好,用小模型可能會卡在複雜任務上。

常見問題 FAQ

Goose 免費嗎?

Goose 本身完全免費開源(Apache 2.0 授權),但你需要自付 LLM API 費用,或使用免費的本地模型如 Ollama。

Goose 支援哪些程式語言?

Goose 是語言無關的 Agent 框架,理論上支援任何程式語言,實際效果取決於底層 LLM 對該語言的支援程度。

Goose 跟 MCP 是什麼關係?

Goose 原生整合 Anthropic 的 Model Context Protocol(MCP),透過 MCP 連接各種開發工具、資料庫和服務。Block 也參與了 MCP 的開發。

用 Goose 安全嗎?會不會誤刪檔案?

Goose 在本地執行,不會把你的 code 送到外部伺服器。但跟所有 AI Agent 一樣,它有自主執行能力,建議在版本控制下使用,隨時可以 rollback。

好不好用,試了才知道。


🇺🇸 Goose Open-Source AI Agent Guide: Block's Local-First Engineering Automation Tool

Goose is an open-source AI agent built by Block (formerly Square) that autonomously executes code installation, editing, testing, and other engineering tasks locally. It supports 15+ LLM providers and has joined the Linux Foundation's Agentic AI Foundation.

What Is Goose and Why Should Developers Care?

If you're still using AI assistants to "suggest" code, Goose wants you to rethink AI's role in your development workflow. Goose isn't just a code completion tool — it's a real AI agent that reads and writes files, runs tests, installs dependencies, and autonomously executes multi-step tasks right in your terminal.

Block's CTO Dhanji Prasanna put it plainly: "Making Goose open source creates a framework for new heights of invention and growth." Block's engineers already use Goose internally to eliminate repetitive work and focus on higher-value tasks.

What Are Goose's Core Features?

  • Autonomous Task Execution: Not just suggestions — Goose searches, navigates, writes code, runs tests, and installs dependencies end-to-end
  • Model Freedom: Supports Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock, and 15+ providers with zero vendor lock-in
  • Native MCP Integration: Connects to 70+ extensions via Anthropic's Model Context Protocol for dynamic system discovery
  • Cross-Platform Native Apps: Desktop apps for macOS, Linux, and Windows, plus full CLI and API access
  • Built in Rust: Core engine written in Rust (58.6% of codebase), delivering strong performance and portability
  • Local-First Execution: Everything runs on your machine — no data leaves your environment, solving privacy and latency concerns

How Does Goose Compare to Claude Code, Cursor, and Copilot?

The AI developer tools space is crowded, but Goose has a clear position:

  • vs Claude Code: Claude Code is Anthropic's terminal AI agent, tied to Claude models. Goose is model-agnostic — plug in any LLM you want
  • vs Cursor: Cursor is a VS Code fork with IDE-integrated AI. Goose is a standalone agent framework with desktop app, CLI, and API interfaces
  • vs GitHub Copilot: Copilot focuses on code suggestions. Goose is action-oriented — it doesn't suggest changes, it makes them

The biggest differentiator? Goose is fully open source under Apache 2.0. You can modify, commercialize, and deploy it freely with no licensing fees.

How to Install Goose

One-line CLI install:

curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash

For the desktop app, head to goose-docs.ai and download the installer for macOS, Linux, or Windows. After installation, select your LLM provider, enter your API key, and you're ready to go.

What Does Goose Joining the Linux Foundation Mean?

In late 2025, the Linux Foundation established the Agentic AI Foundation (AAIF) with three founding projects: Anthropic's MCP, Block's Goose, and OpenAI's AGENTS.md. Platinum members include AWS, Google, Microsoft, Anthropic, OpenAI, and Cloudflare.

This means Goose is no longer just Block's internal tool — it's now foundational infrastructure for the AI agent ecosystem. MCP standardizes how AI connects to tools, Goose provides the open-source agent framework, and AGENTS.md gives AI agents project-specific guidance. Together, they're essentially the "Linux + Docker + README" of the AI agent era.

Real-World Usage Experience

With 40.8k GitHub stars, 4k forks, and 127 releases (latest v1.30.0 on April 8, 2026), community traction is strong. In practice:

  • Setup is straightforward — pick a provider, enter your key, and go
  • The MCP extension ecosystem is rich, with 70+ extensions connecting to various dev tools
  • The Rust core is noticeably fast, feeling snappier than Python-based tools
  • Model flexibility is a real advantage — use Claude, GPT-5.4, or run local models through Ollama

One caveat: agent tool performance depends heavily on the underlying model. Top-tier models like Claude Opus and GPT-5.4 deliver excellent results, while smaller models may struggle with complex tasks.

Frequently Asked Questions

Is Goose Free?

Goose itself is completely free and open source under Apache 2.0. However, you'll need to pay for LLM API costs or use free local models like Ollama.

What Programming Languages Does Goose Support?

Goose is a language-agnostic agent framework that theoretically supports any programming language. Actual effectiveness depends on the underlying LLM's proficiency with that language.

What's the Relationship Between Goose and MCP?

Goose natively integrates Anthropic's Model Context Protocol (MCP) to connect with various development tools, databases, and services. Block also contributed to MCP's development.

Is Goose Safe to Use? Can It Accidentally Delete Files?

Goose runs locally and won't send your code to external servers. But like all AI agents with autonomous execution capabilities, it's recommended to use it under version control so you can always roll back.

You won't know if it's good until you try it.

Sources / 資料來源

常見問題 FAQ

Goose 是免費的嗎?

Goose 本身完全免費開源(Apache 2.0 授權),但需要自付 LLM API 費用,或使用 Ollama 等免費本地模型。

Goose 支援哪些程式語言?

Goose 是語言無關的 Agent 框架,支援任何程式語言,實際效果取決於底層 LLM 的能力。

Goose 跟 MCP 是什麼關係?

Goose 原生整合 Anthropic 的 Model Context Protocol(MCP),透過 MCP 連接開發工具、資料庫和服務。

Goose 跟 Claude Code 有什麼不同?

Claude Code 綁定 Claude 模型,Goose 是模型無關的開源框架,可接任何 LLM 供應商。

用 Goose 會不會誤刪檔案?

Goose 在本地執行,建議搭配版本控制使用,隨時可以 rollback 任何變更。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

ARC-AGI-3 發布:頂尖 AI 全部得分不到 1% | ARC-AGI-3: Every Top AI Model Scored Under 1%

MCP 突破 9700 萬次下載:AI Agent 的「USB-C」為何成為 2026 年最重要的標準? | MCP Hits 97 Million Downloads: Why Model Context Protocol Became the Most Important AI Standard of 2026

LiteLLM 供應鏈攻擊:你的 AI 開發環境可能已被入侵 | LiteLLM Supply Chain Attack: Your AI Dev Environment May Be Compromised