Anthropic's Claude Code is one of the most capable AI coding agents on the planet. It can write, debug, and deploy code autonomously from your terminal. But it costs up to $200 a month, and developers are increasingly fed up with its confusing usage limits. Enter Goose, an open-source alternative from Block (formerly Square) that does much of the same work for exactly zero dollars.
Goose has been picking up serious momentum. The project has racked up more than 26,100 stars on GitHub, with 362 contributors and 102 releases. Version 1.20.1 shipped on January 19, 2026. That's a development pace you'd expect from a well-funded commercial product, not a free open-source tool.
The core difference is where the AI runs. Claude Code sends your queries to Anthropic's cloud servers. Goose can run entirely on your local machine using open-source language models you download yourself through tools like Ollama. No subscription. No usage caps. No code leaving your computer. As Block engineer Parth Sareen put it during a recent livestream, "Your data stays with you, period." He also mentioned using Ollama on planes, which tells you everything about how offline-friendly this setup really is.
To understand why developers are flocking to Goose, look at the Claude Code pricing backlash. The Pro plan ($20/month) limits you to 10 to 40 prompts every five hours. Serious developers burn through that in minutes. Even the $200/month Max plan has restrictions that don't mean what you'd think. Anthropic introduced "weekly hour" limits, but those hours are actually token-based measurements that vary wildly based on your codebase and conversation length. Independent analysis pegged the real limits at roughly 44,000 tokens for Pro and 220,000 for the top-tier Max plan. Some users report hitting daily limits within 30 minutes of intensive work.
Goose is model-agnostic by design, which is its secret weapon. You can connect it to Claude's API, OpenAI's GPT-5, Google's Gemini, or run fully local models like Meta's Llama, Alibaba's Qwen, or DeepSeek's reasoning architectures. It supports the Model Context Protocol (MCP) for connecting to databases, search engines, and third-party APIs. The tool can build entire projects, execute code, debug failures, and orchestrate multi-file workflows autonomously. This isn't just autocomplete. It's a full agentic coding assistant.
Now for the honest trade-offs. Claude 4.5 Opus is still arguably the best AI model for software engineering. One developer who switched to the $200 plan said, "When I say 'make this look modern,' Opus knows what I mean. Other models give me Bootstrap circa 2015." Cloud models also have massive context windows (up to one million tokens for Claude Sonnet 4.5 via API) while most local models top out at 4,096 to 8,192 tokens by default. And cloud inference on dedicated hardware is simply faster than running models on a consumer laptop. You'll also need decent specs to go local. Block recommends 32GB of RAM as a solid baseline, though smaller models can work on 16GB machines.
The bigger picture here matters for anyone building with AI tools. Open-source models are closing the gap fast. Moonshot AI's Kimi K2 and z.ai's GLM 4.5 now benchmark near Claude Sonnet 4 levels, and they're freely available. If that trajectory holds, the quality advantage justifying premium pricing will keep shrinking. Goose sits in a crowded market alongside Cursor ($20 to $200/month), GitHub Copilot, and various other tools, but its combination of full autonomy, model flexibility, local operation, and zero cost carves out a unique position.
For AI-powered professionals, this is a signal worth watching. The era of paying $200 a month for AI coding tools may not last. Whether you switch to Goose today or not, its existence puts real competitive pressure on every paid coding assistant. And for developers who value privacy, offline access, and financial freedom, Goose is already a legitimate option. Both Goose and Ollama are free and open source, available on GitHub and ollama.com respectively.
