Claude Code costs up to $200/month and caps your usage every five hours. Goose does the same thing for free—and runs entirely on your laptop, even on a plane.
The $200 Subscription Rebellion
Anthropic's Claude Code has dominated developer Twitter since New Year's Day. Engineers are posting videos of it building entire apps from a paragraph of text. But there's a catch: the Pro plan ($20/month) gives you just 10-40 prompts every five hours. The Max plans ($100-$200/month) offer more headroom, but developers report hitting daily limits within 30 minutes of real work.
Enter Goose: an open-source AI agent from Block (yes, Jack Dorsey's payments company) that runs locally on your machine. No subscription. No rate limits. No cloud dependency. It's exploded to 26,100 GitHub stars in months, and it's become the go-to alternative for developers who want control over their AI workflow.
What Goose Actually Does
Goose is a command-line tool (or desktop app) that autonomously writes, debugs, and executes code. It can build projects from scratch, orchestrate workflows across multiple files, interact with APIs, and run tests—all without constant hand-holding. Unlike code completion tools like GitHub Copilot, Goose takes full tasks and completes them end-to-end.
The killer feature? It's model-agnostic. You can connect it to Claude, GPT-5, or Gemini if you have API access. Or—here's where it gets interesting—you can run it entirely locally using tools like Ollama, which let you download open-source models and execute them on your own hardware. Your code never leaves your machine.
One developer put it bluntly during a livestream: "I use Ollama all the time on planes—it's a lot of fun!" No internet required. No usage caps. Just you and the AI, offline.
How to Set Up Goose with a Local Model in Three Steps
Step 1: Install Ollama. Download it from ollama.com. Once installed, pull a coding model with one command: ollama run qwen2.5. The model downloads automatically and starts running locally.
Step 2: Install Goose. Grab the desktop app or CLI from Goose's GitHub releases page. Pre-built binaries are available for macOS, Windows, and Linux.
Step 3: Connect them. In Goose Desktop, go to Settings → Configure Provider → Ollama. Set the API host to http://localhost:11434 and submit. For the CLI, run goose configure and follow the prompts.
That's it. You're now running an AI coding agent on your laptop with zero ongoing costs.
The Trade-Offs You Need to Know
Goose with a local model isn't a perfect substitute for Claude Code. Claude Opus 4.5 remains arguably the best AI for software engineering—it excels at understanding complex codebases and producing high-quality code on the first attempt. Open-source models have improved dramatically, but a gap persists.
You'll also need decent hardware. Block's documentation suggests 32GB of RAM as a baseline for larger models. Smaller models can run on 16GB, but entry-level laptops with 8GB will struggle. And local models process requests more slowly than cloud-based services running on dedicated server hardware.
But here's what you gain: complete privacy (your conversations never leave your machine), zero subscription fees, no rate limits, and the ability to work offline. For many developers, that's a worthwhile trade.
What This Means for Learners
If you're learning AI or building AI skills, Goose is a masterclass in what's possible when you control your tools. It's also a forcing function: to use it effectively, you need to understand how language models work, how to configure local environments, and how to think about trade-offs between model quality, speed, and cost.
Want to go deeper? Our Vibe Coding with Cursor and Windsurf course covers AI-assisted development workflows in depth. Or explore Claude Code: Ship Without Chaos to understand how the paid alternative works—and when it's worth the money.
The $200-a-month era for AI coding tools may be ending. Open-source models are improving fast. The gap is narrowing. And tools like Goose prove that you don't need a subscription to build real software with AI.
