01What is Qwythos 9B and how it works

Qwythos 9B is a free, open-source AI model that runs entirely on your own computer. It is designed to give you Claude-style creative reasoning without needing the cloud or paying for tokens.

In this episode, Julian Goldie explains that Qwythos 9B is built on a Qwen 3.5 9B base and is available through Ollama. It uses a million-token context window, which is huge, and can reason and call tools, making it good for agentic tasks.

The model is small—only 5.6 GB—so it can run on a Mac laptop. Julian tested it on a Mac Studio and found it can create simple apps like a digital clock, a to-do list, and even a snake game.

02Key features of Qwythos 9B

Qwythos 9B comes with several standout features that make it useful for local AI work.

  • 1M token context window—officially supports a huge amount of context, but note that real-world use may be limited by your hardware.
  • Claude-style reasoning—trained to think creatively and logically, similar to frontier models but smaller.
  • Tool calling—native function calling, which makes it suitable for agentic systems.
  • Free and private—no cloud, no token costs, and nothing leaves your computer.
  • Lightweight—only 5.6 GB in size, about a quarter of Gemma 12B.
  • Local execution—works offline; you can use it even without internet.

These features make it appealing for developers and hobbyists who want to experiment with local AI.

03How to install Qwythos 9B with Ollama

Getting Qwythos 9B running on your machine is simple if you follow these steps. Julian showed exactly how to do it in the episode.

  1. Make sure you have Ollama installed on your computer. If not, download it from the official site.
  2. Open your terminal.
  3. Run the command provided by Ollama to pull the Qwythos 9B model.
  4. Once the download is complete, run the model using the Ollama run command.
  5. Start interacting with the model in the terminal or plug it into your own agent system.

Julian also mentioned that he wired Qwythos 9B into his agent OS local engine, which allows him to test it in a more integrated way. But for a simple start, the terminal is enough.

04How to use Qwythos 9B: examples shown

Julian demonstrated several practical uses for Qwythos 9B in the episode. He tested it by building six small apps, and most worked well.

  • To-do list app—you can add tasks and clear the list, all running locally.
  • Digital clock—a simple clock that runs on your machine.
  • Snake game—a classic game that works, though it might be a bit glitchy.
  • Calculator—a functional calculator app.
  • Landing page design—Qwythos 9B created a decent design, surprisingly better than some frontier models.

These examples show that Qwythos 9B can handle basic coding tasks and creative projects. It's not perfect—some outputs were messy—but it works for small projects.

05Results and comparison with Ornith 9B

Julian compared Qwythos 9B with another local model, Ornith 9B. He shared his observations about speed and performance.

In his tests, Qwythos 9B came back twice as fast as Ornith 9B for the same work. However, Julian noted that Ornith 1.0 is generally outperforming Qwythos 9B in terms of quality.

He also tested the context window. He found that despite the claimed 1 million tokens, his setup sometimes had a tiny window because Ollama loads it with a much smaller window by default. The actual size depends on your memory.

Overall, the results show that Qwythos 9B is a solid local model for its size, but it's not at frontier level.

06Tips, limits, and mistakes to avoid

Using Qwythos 9B has its quirks. Julian shared some tips and warnings from his testing.

  • Context window size matters—the 1M token window doesn't work out of the box. You need to pick a window that fits your RAM, or you may run out of tokens.
  • It can be slow to load—the model may take time to start up, depending on your hardware.
  • Expect occasional glitches—sometimes the model cuts off responses or produces messy output, but other times it works perfectly.
  • No built-in tools or memory—it's just a model, so you need to add your own tools if you want them.
  • Choose the right model size—Ollama offers different versions; smaller ones are faster but less sharp, larger ones are more accurate but heavier.

Keep these points in mind to get the best experience with Qwythos 9B.

07Who should use Qwythos 9B and why it matters

Qwythos 9B is for anyone who wants to run AI locally without paying for cloud services. It's great for developers, hobbyists, and people who care about privacy.

Julian emphasized that local models like this are getting better and better. While they're not yet at frontier level, they are fun to experiment with and can handle real tasks.

The fact that it's free and private is a big deal. You can use it even without internet. Plus, it's small enough to run on a laptop, making it accessible to many people.

If you're into building AI agents or just want to try out a local model, Qwythos 9B is worth checking out.

08How Qwythos 9B fits into an agent OS

Julian showed how you can plug Qwythos 9B into a larger system, like the agent OS he uses. This allows the model to act as the engine for your agents.

He has a setup where you can switch between local models easily, preview apps you've built, and use free cloud code. The agent OS also includes tools like Hermes Jarvis, Hermes Oracle for news, a workspace, and a lead generation tool.

Julian mentioned that a local model running as the engine of your whole agent OS is powerful. It's free, private, and can run with whatever you want.

He also mentioned that you can use Qwythos 9B with other harnesses like Hermes, though he hasn't tested that combination yet.