Block / Linux Foundation · Open Source

Goose
AI Agent

An autonomous AI agent that runs on your machine — it writes code, fixes bugs, manages files, calls APIs and orchestrates full workflows. No subscription. Works with any LLM. Scroll to AI-Web for the step-by-step playbook to build a website with AI.

Free · Apache 2.0 Built in Rust Any LLM
27K+
GitHub stars in 12 months
3,000+
MCP extension servers
15+
LLM providers supported
5,000
Block staff use it weekly
01 — The Idea

What is Goose?

Think of Goose as a digital employee on your laptop. You tell it what you want in plain language — and it figures out the steps, runs commands, fixes errors, and delivers results.

Built by Block (formerly Square, founded by Jack Dorsey) and released in January 2025. In December 2025 it was contributed to the Linux Foundation's Agentic AI Foundation — alongside Anthropic's MCP and OpenAI's AGENTS.md — making it global infrastructure, not a proprietary product.

It runs in three ways: a desktop app, a command-line tool (CLI), or via an API you can embed in your own products.

CapabilityChatGPT🪿 Goose
Suggests code fixes
Actually runs the code
Reads your files
Installs dependencies
Connects Slack / Drive / GitHub
Works with any AI model
Runs on your machine (private)
Free to use
02 — Mechanics

How it works — step by step

1

You type a task in plain English

Open the Goose desktop app, CLI, or connect via API and type what you want — no code required. Goose understands natural language just like a human assistant would.

"Set up a new JavaScript project with Express and MongoDB" "Find all bugs in my code and fix them"
Natural languageNo coding needed
🪿Goose · Desktop
🧑 Set up a new JavaScript project with Express and MongoDB, add a health-check route and a README
Send ↵
2

Goose sends your request to an AI brain (LLM)

Your instruction goes to a Large Language Model — the engine that understands and thinks. You choose the brain: Claude, GPT, Gemini, or a free local model on your own computer.

ClaudeGPTGeminiOllama (local)

💡 With a local model, no data leaves your computer — 100% private.

🧠Settings · Choose your AI brain
Claude (Anthropic) — Sonnet✓ active
GPT-4o (OpenAI)
Gemini (Google)
Llama 3 · Ollama local · private
3

The AI plans the steps

The LLM doesn't just answer — it builds an action plan, deciding which tools to use and in what order.

A. Read the project files · B. Run the test suite · C. Identify failures · D. Write code to fix them · E. Re-run tests to confirm

🧭Goose · Planning
# Goose drafts an action plan before touching anything Plan ── 5 steps 1. Scaffold project & install express, mongoose 2. Create server.js + health-check route 3. Run the test suite 4. Fix any failures, write the README 5. Re-run tests to confirm
4

Goose calls the right tools (extensions)

Goose uses the Model Context Protocol (MCP) — an open standard — to connect to 3,000+ tools. Each tool is a specialist worker Goose can hire for a specific job.

📁 Files & Shell🐙 GitHub💬 Slack📊 Drive🐳 Docker🗄️ Databases
🔌Goose · MCP tool calls
# Goose hires the right specialist tool for each job ▸ shell npm init -y && npm i express mongoose ▸ files write server.js, routes/health.js ▸ shell npm test ▸ github open PR #7 "add health-check"
5

Tools execute — results come back

Each tool does its job and reports back. If something fails, Goose retries automatically — like a smart employee who troubleshoots without bothering you.

Example: Goose asks GitHub to list open issues → gets 12 → reads them → fixes issue #7 → submits a pull request. All automatically.

⚙️Goose · Execution log
✓ packages installed (1.8s) ✓ server.js created · health route mounted ⚠ test failed — port 3000 already in use ↻ Goose retries automatically → switch to PORT env ✓ npm test — 4 passing, 0 failing ✓ PR #7 opened on GitHub
6

Goose reports back to you

When done, Goose summarises what it did, what it changed, and what needs your attention. You stay in control and can redirect at any point.

✅ Task complete📋 Action summary🔁 Continuous loop
📋Goose · Summary
✓ Task complete · 58s
📦Created Express + MongoDB project — 9 files, dependencies installed.
🧪Tests: 4 passing, 0 failing (1 issue auto-fixed).
🐙Opened PR #7 "add health-check route" on GitHub.
👉Needs you: review & merge the PR, then set your MONGO_URI.
03 — The Map

Goose ecosystem — mind tree

🪿 Goose
AI Agent
🧠 AI BrainsChoose your LLM
🤖Claude (Anthropic)
💬GPT (OpenAI)
💎Gemini (Google)
🦙Llama (local)
🔓Ollama (free local)
☁️Azure / Bedrock
🔌 ExtensionsMCP connections
🐙GitHub
💬Slack
📁Google Drive
🗄️Databases (SQL)
🐳Docker
📝Notion
🌐Web browser
📮Email / Gmail
⚙️ Built-in ToolsAlways available
📂Read / write files
💻Run shell commands
🧪Execute tests
📦Install packages
🔍Search code
🌐Call web APIs
📋 RecipesSaved workflows
📊Weekly status report
🐛Auto bug fix
🚀Deploy pipeline
📈Data dashboard
🔄Scheduled (cron)
🤝Share with team
🖥️ InterfacesHow you use it
🖥️Desktop app
⌨️CLI (terminal)
🔗API (embed anywhere)
🦺Claude Desktop (MCP client)
04 — In Practice

What can you do with Goose?

💻

Coding & development

Set up entire project environments, write, test and debug code, run test suites and auto-fix failures, commit and push to GitHub — in one prompt.

🐛

Auto bug fixing

Paste an error. Goose reads your code, finds the root cause, writes the fix, re-runs tests and confirms it's resolved — without you typing a line.

📊

Data analysis

Point Goose at a CSV or database and ask "build me an interactive dashboard." It reads the data, writes the chart code and serves it live in your browser.

🗂️

File & system tasks

Rename 500 files by pattern, reorganise folders, extract data from PDFs, merge spreadsheets — any repetitive file task that would take you hours.

🔗

Workflow automation

Connect GitHub + Slack + Jira. When a bug is filed, Goose reads it, tries to fix it, then posts in Slack with a PR link. Zero human steps.

🔒

Private / air-gapped

Run with a local model (Ollama). Zero data leaves your machine — perfect for sensitive business data, healthcare records or classified environments.

05 — Reusable

Recipes — reusable workflows

A Recipe is a saved shortcut — a pre-written set of instructions Goose follows every time. Write it once, share it with your team, and run it with one command. Think of it as a menu item: the same steps, perfectly, every time.

Recipe · Weekly Status Report
1
Trigger

Run every Monday at 9am automatically via cron schedule.

2
Action · Read git history

Goose reads all code commits from the past 7 days.

3
Action · Check open issues

Pulls all open bugs and tasks from GitHub.

4
Output · Post to Slack

Formats a clean report and posts to #engineering.

# recipe.yaml
title: Weekly Status Report
schedule: "0 9 * * MON"
extensions: [developer, github, slack]
instructions: |
  Analyze git history last 7 days.
  Summarize commits + open issues.
  Post report to #engineering.

Real example: a developer typed one prompt — "Read my blog CSV files in /data, analyze them and create an interactive dashboard" — and Goose read the files, wrote the code and served a live dashboard. No coding needed.

Real Goose session
YOU"Set up a new JS project with Express, Mongoose, Nodemon"
GOOSECreates folder structure, installs all packages, generates a README with latest versions, initializes Git — in 60 seconds
YOU"Now fix any bugs you find"
GOOSERuns tests, finds 2 issues, writes fixes, re-runs tests, confirms ✅ — reports back with a summary
Why this matters for non-coders

Goose isn't just for programmers. Any business person can use it to automate repetitive file tasks, pull data from multiple sources into one report, connect business tools (CRM, email, spreadsheets) and schedule recurring reports automatically.

You describe the outcome. Goose figures out the how.

06 — The Field

Goose vs alternatives

🪿 Goose

Free · Open Source
Cost$0 (free)
LLM choiceAny (15+ providers)
Data privacy100% local option
Extensions3,000+ MCP servers
Runs code?Yes ✓
Vendor lock-inNone

🔵 Claude Code

Paid subscription
Cost$20–$200/mo
LLM choiceClaude only
Data privacyCloud
ExtensionsMCP (growing)
Runs code?Yes ✓
StrengthTop-tier accuracy

💬 ChatGPT

Paid subscription
Cost$20/mo
LLM choiceGPT only
Data privacyCloud
ExtensionsLimited
Runs code?Sandboxed
Access your files?No ✗
The bottom line

Goose is the only free, open-source AI agent that runs on your own machine, works with any AI model, connects to 3,000+ tools, and has been adopted by the Linux Foundation as global infrastructure. For businesses that need power without vendor lock-in, it's the professional choice.

Free
Apache 2.0 license
Private
Runs 100% locally
Open
Any LLM, any tool
Powerful
3,000+ integrations
07 — AI-Web

Build a website with AI — the toolkit & the steps

You don't need an agency or a CS degree to ship a professional website. With an AI agent like Goose (or Claude Code) you describe the outcome and the agent does the building. Here is the exact toolkit and the step-by-step method used to build pages like this one — on the EatCookJoy / Aziz Saif house theme.

🧰 Tools to use

🪿AI BuilderGooseAutonomous agent that scaffolds, writes and edits the whole site from plain-English prompts.Free
🤖AI Pair BuilderClaude / Claude CodeGenerates clean HTML/CSS, refactors, reviews and explains — your design & code co-pilot.
📝EditorVS Code / CursorWhere you and the agent see and edit the files. Cursor has AI baked in.Free
🎨Design / MockupFigma / v0Sketch layout and mood, or generate a first UI to hand to the agent.Free tier
🐙Version + HostGitHub + PagesStores your code and hosts the live site for free at username.github.io or a custom domain.Free
🌐Domain + CDNCloudflare / NamecheapBuy your domain, point DNS, get free SSL and a global CDN in front of the site.
🔤TypographyGoogle FontsPlayfair Display, Inter and Space Grotesk — the house theme fonts, loaded free via CDN.Free
📈Analytics + SEOPlausible / Search ConsoleSee traffic privately and confirm Google indexes your pages.

Minimum viable stack to ship for free: Goose + VS Code + GitHub Pages + a domain. Everything else is an upgrade.

📚 Deep-dive guides

🎬 DIY-Web · AI Video Claude Code + Higgsfield — 100+ AI videos in minutes The full step-by-step setup — Node.js, Claude Code, Playwright MCP & Higgsfield — to generate UGC ad videos at scale. Gulf-ready, halal-appropriate.

🪜 Step-by-step instructions

1

Write a one-page brief

Before any tool, decide three things: who it's for, what the one job of the page is, and the single call-to-action. Two sentences is enough — the agent builds far better with a clear target.

AudienceOne jobOne CTA
2

Pick your stack & install the tools

Install the agent and an editor, then create an empty project folder. For a static site (perfect for portfolios, explainers and landing pages) plain HTML + CSS is all you need — fast, durable and free to host.

brew install block-goose-cli code my-site
3

Lock the design system first

Decide colours, fonts and spacing before building pages — this is what keeps every page on-brand. Reuse the house theme below (cream paper, terracotta, gold & sage, Playfair + Inter + Space Grotesk). Tip: this exact theme is saved as a reusable Claude Code skill so the agent applies it automatically every time.

Theme tokens: cream #FBF7F1 · terracotta #C45A3D · gold #D9A86C · sage #7A8C5C · ink #16110D — fonts: Playfair Display (display), Inter (body), Space Grotesk (labels).

4

Scaffold with one prompt

Hand the brief and the design tokens to the agent and let it generate the first full page. Then review it live and refine section by section in plain English.

Starter prompt → Goose / Claude Code # Paste this into the agent
Build a one-page static website at index.html.
Audience: [who]. Goal: [one job]. CTA: [button].
Use the EatCookJoy / Aziz Saif house theme:
  colours cream #FBF7F1, terracotta #C45A3D, gold #D9A86C, ink #16110D;
  fonts Playfair Display (headings), Inter (body), Space Grotesk (labels);
  editorial layout, grain overlay, pill buttons, soft cards, reveal-on-scroll.
Sections: hero, what, how it works, CTA. Make it responsive & accessible.
5

Write the content like an editor

Replace placeholder copy with real, specific words. Lead with the outcome, keep sentences short, use one clear heading per section. Ask the agent to tighten and proofread — but you own the message.

Outcome-firstShort sentencesOne idea / section
6

Make it responsive & accessible

Check the page on a phone width, confirm tap targets are big enough, add alt text to images, keep colour contrast high, and make sure it works without JavaScript. Ask the agent: "audit this page for mobile and accessibility and fix it."

Mobile-firstAlt textContrastKeyboard nav
7

Optimise for search & AI engines (SEO · AEO · GEO)

Add a descriptive <title> and meta description, Open Graph tags, a canonical link, JSON-LD schema, a sitemap.xml and robots.txt. Clear headings and direct answers help you show up in Google and in AI answers (Perplexity, ChatGPT Search, Gemini).

Meta + OGJSON-LD schemaSitemapAnswer-first copy
8

Connect the domain & deploy

Push to GitHub, enable Pages, and point your domain at it with a CNAME file. The agent can run every command for you. Your site is now live, on HTTPS, for free.

Deploy → GitHub Pages # one-time setup
git init && git add . && git commit -m "launch site"
git branch -M main
git remote add origin git@github.com:you/your-site.git
git push -u origin main
# add a CNAME file containing your domain, then
# enable Pages in repo Settings → Pages → branch: main
9

Measure, iterate & keep it on-brand

Add analytics, watch what visitors do, and improve the weak sections. Because the theme lives in a reusable skill, every new page you add stays visually consistent — your site grows without drifting off-brand.

AnalyticsIterateConsistent themeShip weekly

⚡ The stack at a glance

JobToolWhyCost
Build the siteGoose / Claude CodeTurns plain-English prompts into clean, working codeFree
Edit & previewVS Code or CursorSee files, preview live, accept agent editsFree
Version controlGit + GitHubHistory, backup, and the deploy pipelineFree
HostingGitHub PagesFree static hosting with HTTPS and a custom domainFree
Domain + CDNCloudflare / NamecheapCustom URL, free SSL, fast global delivery~$10/yr
FontsGoogle FontsPlayfair Display · Inter · Space GroteskFree
SEO / AI searchSearch Console + schemaGet indexed by Google and cited by AI enginesFree
AnalyticsPlausible / GA4Understand visitors and improveFree / Paid

Want a website that builds itself — on brand, every time?

This page was generated with an AI agent on the EatCookJoy / Aziz Saif house theme. The same theme is saved as a reusable skill, so every future page stays perfectly consistent.

Explore azizsaif.com