Imagine ditching expensive AI coding tools and embracing a completely free, local alternative. Sounds too good to be true, right? Well, that's exactly what I set out to explore by testing a combination of Goose, an open-source agent framework, and Qwen3-coder, a powerful coding-focused language model. But here's where it gets controversial: Can these free tools truly replace pricey options like Claude Code? Let's dive in and find out.
Jack Dorsey, the mastermind behind Twitter (now X), Square (now Block), and Bluesky, sparked curiosity in July with a cryptic tweet: "goose + qwen3-coder = wow." This simple equation ignited interest in both tools, leaving many wondering if they could indeed create a free rival to Claude Code. Goose, developed by Block, is an open-source agent framework, while Qwen3-coder, from Alibaba, is a coding-centric large language model. Both are free, but their combined potential is what has the tech world buzzing.
And this is the part most people miss: While the setup is straightforward, it does require a robust local machine. I decided to put this duo to the test, starting with the installation process. Here's a step-by-step guide to getting everything up and running:
- Downloading the Software: Begin by downloading Goose from GitHub and Ollama from its official website. Ollama serves as the LLM server, and you'll later download the Qwen3-coder model through it.
- Installing Ollama and Qwen3-coder: Install Ollama first, preferably the app version for ease of use. Once installed, you'll see a chat-like interface. Select Qwen3-coder:30b from the model list—a coding-optimized model with 30 billion parameters. Note that the 17GB model downloads only when prompted, so ensure you have sufficient storage. This local setup is a game-changer, as it keeps your AI operations entirely on your machine, away from the cloud.
- Configuring Ollama: Make Ollama visible to other applications by enabling 'Expose Ollama to the network' in the settings. I opted for a 32K context length, given my machine's 128GB RAM, but you can adjust this based on your hardware.
- Installing Goose: Run the Goose installer, choosing the version compatible with your system (e.g., MacOS Apple Silicon Desktop). Upon first launch, configure Goose to connect to Ollama by selecting it from the 'Other Providers' section and choosing the Qwen3-coder model.
With everything set up, I put Goose to the test using my standard challenge: building a simple WordPress plugin. Here's the surprising part: It took five attempts for Goose to get it right. While this might seem like a setback, there's a silver lining. Unlike chatbot coding, agentic coding tools like Goose and Claude Code work directly on the source code, meaning each correction improves the codebase. This iterative process, though time-consuming, can lead to more refined results.
Performance-wise, my M4 Max Mac Studio with 128GB RAM handled the setup admirably, even with multiple resource-intensive apps running simultaneously. The turnaround time for prompts was comparable to cloud-based alternatives like Claude Code and OpenAI Codex, which is impressive for a local setup.
But here's the burning question: Can this free, local solution truly replace expensive cloud-based tools? While my initial impressions are positive, a more comprehensive evaluation is needed. I plan to run a larger project through this setup to determine its long-term viability. Stay tuned for that in-depth analysis.
Now, I want to hear from you: Have you experimented with local coding LLMs like Goose, Ollama, or Qwen3-coder? How did your setup experience compare? If you've used cloud options, how do you think local performance stacks up? Share your thoughts and experiences in the comments below. Let's spark a discussion and explore the future of AI coding together.
For more updates on my projects, follow me on social media and subscribe to my weekly newsletter. Let's keep the conversation going!
