Goose gives you Claude Code-level AI coding for free — no subscription required

The AI coding revolution has a dirty little secret: it’s expensive as hell.

<a href="https://video.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code from Anthropic is genuinely impressive — it writes, debugs, and deploys code right from your terminal. But the pricing structure has been pissing off developers since day one. Between $20 and $200 a month, plus rate limits that reset every five hours and a token system that nobody seems to fully understand, it’s starting to feel like the old guard of SaaS monetization all over again.

Enter Goose. It’s an open-source AI agent built by Block (yeah, the Jack Dorsey payments company formerly known as Square). It does basically the same thing as Claude Code, but runs entirely on your local machine. No subscription. No cloud dependency. No arbitrary caps that kick in when you’re in the zone.

“Your data stays with you, period,” said Parth Sareen, a software engineer who demoed the tool during a livestream. That line sums up the whole appeal: full control, including the ability to work offline — even on a damn airplane.

Goose has been quietly exploding. It’s sitting at over 26,100 stars on GitHub, with 362 contributors and 102 releases since launch. The latest version, 1.20.1, dropped on January 19, 2026. That’s a development cadence that rivals any commercial product.

For developers who are tired of Anthropic’s pricing games, Goose is a genuinely free, no-strings-attached alternative for serious work.

The Claude Code rate limit mess

To understand why Goose matters, you need to appreciate just how badly Anthropic botched their pricing.

The free plan for Claude Code gives you zero access. The Pro plan is $17/month if you pay annually ($20 monthly), and it limits you to 10 to 40 prompts every five hours. If you’re doing any real work, you’ll burn through that in minutes.

The Max plans — $100 and $200 per month — give you more headroom: 50 to 200 prompts and 200 to 800 prompts respectively, plus access to Claude 4.5 Opus. But even those come with restrictions that have the developer community fuming.

In late July, Anthropic introduced weekly rate limits. Pro users get 40 to 80 hours of Sonnet 4 usage per week. Max users at the $200 tier get 240 to 480 hours of Sonnet 4, plus 24 to 40 hours of Opus 4. Nearly five months later, people are still pissed.

The problem? Those “hours” aren’t actual hours. They’re token-based limits that vary wildly depending on your codebase size, conversation length, and code complexity. Independent analysis suggests the real per-session limits are roughly 44,000 tokens for Pro users and 220,000 for the $200 Max plan.

“It’s confusing and vague,” one developer wrote in a widely shared analysis. “When they say ’24-40 hours of Opus 4,’ that doesn’t really tell you anything useful about what you’re actually getting.”

The backlash on Reddit and developer forums has been fierce. Some users report hitting their daily limits within 30 minutes of intensive coding. Others have canceled subscriptions entirely, calling the new restrictions “a joke” and “unusable for real work.”

Anthropic has defended the changes, saying the limits affect fewer than five percent of users and target people running Claude Code “continuously in the background, 24/7.” But they haven’t clarified whether that’s five percent of Max subscribers or five percent of all users — a distinction that matters a lot.

How Block built a free AI coding agent that works offline

Goose takes a completely different approach. It’s what engineers call an “on-machine AI agent.” Instead of sending your queries to Anthropic’s servers, Goose runs entirely on your local computer using open-source language models that you download and control.

The project’s documentation describes it as going “beyond code suggestions” to “install, execute, edit, and test” code autonomously. In practice, that means you can give it high-level instructions — “refactor this module to use async/await” or “add error handling to all API routes” — and it will figure out the implementation, run tests, and iterate until it works.

Because it’s local, there’s no latency from network calls, no data leaving your machine, and no rate limits. You can work on a plane, in a coffee shop with spotty Wi-Fi, or in an air-gapped environment. That’s a big deal for anyone dealing with sensitive codebases or compliance requirements.

Goose supports multiple models. You can use GPT-4o, Claude 3.5 Sonnet, or any open-source model you’ve downloaded. The default setup uses smaller, faster models for simple tasks and scales up to larger ones when needed. It’s not as polished as Claude Code out of the box — you’ll need to configure your model provider and possibly download weights — but the trade-off is worth it for many developers.

I’ve been testing Goose for about two weeks now. The setup took maybe 20 minutes, and that included downloading a 7B parameter model. Performance is surprisingly good for routine tasks like writing boilerplate, fixing lint errors, and generating tests. It struggles with complex architectural decisions that require deep context, but so does Claude Code when you’re on the Pro plan and hitting limits every half hour.

The real win is the pace of development. Goose has 362 contributors pushing updates regularly. Features that Anthropic gates behind pay tiers — like multi-file editing, autonomous debugging, and integration with version control — are available in Goose for free. The community is active on Discord and GitHub, and the documentation is solid.

The bigger picture: local AI is the future

Goose isn’t just a Claude Code clone. It’s part of a broader shift toward local, open-source AI tools that give developers control back. We’ve seen it with Ollama for running models locally, with Continue.dev for code completion in IDEs, and now with Goose for autonomous agents.

The appeal is obvious: no vendor lock-in, no surprise bills, and no data leaving your machine. For a generation of developers who’ve watched cloud costs spiral out of control, local AI feels like a breath of fresh air.

There are downsides, of course. You need a decent machine to run larger models. Goose won’t match the raw power of Claude 4.5 Opus running on Anthropic’s infrastructure. And if you’re working on a massive codebase, you’ll still benefit from the cloud-based context windows that Claude Code offers.

But for day-to-day development work — the kind where you’re writing tests, refactoring code, fixing bugs, and generating documentation — Goose is more than capable. And it doesn’t cost a dime.

Anthropic is betting that developers will pay a premium for convenience and power. Block is betting that the open-source community can build something good enough for free. Based on what I’ve seen so far, the second bet is looking pretty smart.