01What Is Free Claude Code and How It Works

Free Claude Code is a way to use the real Claude Code command-line interface (CLI) without paying for tokens. Instead of using Anthropic's paid model, you point the CLI to a free local model like Google's Gemma 4. This lets you run coding agents for free, 24/7, on your own machine.

The key update is Gemma 4 from Google, which is now 90% faster on Apple Silicon when run with Ollama using MLX. This speed boost makes local models practical for coding tasks. You can also use Gemma 4 via a free API from OpenRouter if you don't have a Mac.

In this episode, Julian Goldie shows how he plugged this setup into his Agent Operating System. You get the same Claude Code CLI you know, but every token is free. Your code stays local and private.

02Main Features: What You Get with Free Claude Code

Free Claude Code gives you several important features:

  • Real Claude Code CLI – It's not a watered-down version. It's the same tool, but pointed at a different model.
  • Free tokens – No more paying per token or hesitating before big tasks.
  • Local and private – Your code stays on your machine, not in the cloud.
  • 24/7 autonomous operation – Let it run in the background while you do other things.
  • Multiple ways to run – Use it locally with Ollama and MLX, or via a free API from OpenRouter.
  • Works with Agent OS – Julian has integrated it into his agent operating system, including a local engine.

These features make free Claude Code a practical choice for simple coding tasks, blog posts, and landing pages.

03How to Set Up Free Claude Code: Step-by-Step

Setting up free Claude Code is simple, says Julian. Here's how to do it:

  1. Download Ollama – It's free and available for macOS and other platforms.
  2. Download Gemma 4 – This is Google's free local model, also available through Ollama.
  3. Install Claude Code – You get the open-source Claude Code CLI, which is free.
  4. Point Claude Code to the local model – Use a single terminal command to switch Claude Code to Ollama with Gemma 4 MLX.
  5. Start coding – Run your first task and see it work.

If you don't have a Mac or Apple Silicon, you can use OpenRouter instead. Go to OpenRouter, type "free" in the search, find Gemma 4, and get a free API key. Then plug that API into free Claude Code.

04How to Use Free Claude Code in Real Projects

In the episode, Julian demonstrates several ways to use free Claude Code. First, he shows his Agent Factory, which runs with Gemma 4. You can preview what the agent has built and ask it to create new things.

For example, he asks it to "create a to do list app" and hits build. The agent works in the background and completes the task.

Second, you can use it inside the workspace. He goes to the chat and says, "build out a Space Invaders game." Free Claude Code starts coding in the background. It's not instant, but you can come back later and see the result.

Third, you can use it with the Agent OS local engine. This doesn't even require Claude Code. You ask the local engine to build something, preview the output, and it goes into your workspace.

You can also run tasks on a schedule. Since it's autonomous, you can set it to run for two or three hours and check back when it's done.

05Results and Comparisons: Speed and Quality

Julian highlights that Gemma 4 is 90% faster on Apple Silicon with Ollama using MLX. This makes local models more usable than before. He also compares the old way with the new way:

  • Old way – Every Claude session cost tokens. You hesitated before big tasks. Your code went to the cloud. Local models were too slow, so you avoided them.
  • New way – Free Claude Code with Gemma 4 gives you the real CLI, free tokens, local privacy, and fast performance.

He also tested Hermes agent with Gemma 4. The note is that Hermes uses a lot of tools, which makes Gemma 4 slow. But if you run tasks in the background or on a schedule, speed doesn't matter as much.

You can just wait.

For quality, Julian says local models are good for simple tasks, not complex ones. He has a local leaderboard called Goldie Bench where he tests models on 42 different tasks.

06Tips, Limits, and Mistakes to Avoid

Julian shares several tips and warnings:

  • Don't use local models for super complicated tasks – Use a frontier model instead. For a cheap option, he mentions GLN 5.2.
  • Great for basic tasks – Writing a blog post or creating a landing page is fine with Gemma 4.
  • Be patient with Hermes – It uses many tools, so it will be slow. But that's OK for background tasks.
  • You don't need to be technical – Julian says non-technical people in his community use Agent OS successfully.
  • Test models yourself – Don't take his word for it. Check Goldie Bench and see what you think.
  • Own your system – Relying on closed models is risky. If they get taken down, your workflow breaks. Local models are safer.

Avoid the mistake of expecting frontier-level performance from a local model. Choose the right model for the task.

07Who Is Free Claude Code For and Why It Matters

Free Claude Code is for anyone who wants to use AI coding agents without spending money on tokens. It's also for people who care about privacy and want their code to stay local. Julian explains that owning your system is increasingly important because closed models can be taken down at any time.

This setup is especially useful for non-technical users. Julian notes that many people in his community use the Agent OS and get great results. He personally answers questions daily, and there's a community for support.

By using free Claude Code, you can run agents 24/7 without worrying about costs. You can build simple apps, write content, and automate tasks. As local models keep improving, this approach becomes more powerful.

Julian plans to test more models and update his leaderboard.

08More Ways to Run Free Claude Code: API and Agent OS

Besides local setup, free Claude Code works with an API. If you don't have a Mac or Apple Silicon, you can use OpenRouter. There, you can find Gemma 4 available for free.

Get a free API key, plug it into free Claude Code, and you're set. This way, you still get the agent harness without paying.

Julian also integrated free Claude Code into his Agent Operating System. This system includes a memory component and works with multiple agents. You can use it with the local engine, which is another way to run without Claude Code itself.

He emphasizes that the setup is simple. One terminal command switches Claude Code to Ollama with Gemma 4 MLX. And you can choose the best model for your task.