01What is Laguna XS 2.1?

Laguna XS 2.1 is a new, free, and open-source local AI model made by a company called Poolside. It is designed for agentic coding and terminal work. This means it helps you build software by following instructions, like creating apps or landing pages.

The model just got updated and is available on Hugging Face. It is lightweight, so it runs fast on your own computer. You can use it offline and keep your data on your machine.

It also has a 256K context window, which is enough for most local tasks.

02Main Features of Laguna XS 2.1

Laguna XS 2.1 comes with several key features that make it useful for local AI work:

  • Free and open-source, so anyone can download and use it.
  • Designed specifically for coding and agentic tasks.
  • 256K context window for handling larger projects.
  • Can run locally on your machine, offline, with no cloud dependency.
  • Also available on OpenRouter with a free API key.
  • Works well for building simple landing pages and mini apps.

In the episode, Julian tested it on a Mac Studio M4 Max with 36GB memory, and it ran smoothly. The model is not great for 3D models, but it handles standard coding tasks well.

03How to Set Up Laguna XS 2.1

Setting up Laguna XS 2.1 is simple, and you have a few options. Here is how you can get started:

  1. Download the model from Hugging Face, where it is available for free.
  2. Install it on your local setup, like the Agent OS, by plugging it into the local engine.
  3. Or use it inside FreeClaud Code, selecting Laguna XS 2.1 as the model.
  4. Alternatively, set up a separate Hermes agent profile with this local model.
  5. If you prefer not to run local, go to OpenRouter and get a free API key.
  6. In the terminal, type 'Hermes model' to select OpenRouter and enter your API key.

Once set up, you can start building right away. Julian mentioned that everything you create is saved in the workspace, making it easy to manage.

04Using Laguna XS 2.1 for Coding

In the episode, Julian showed three examples of what he built with Laguna XS 2.1:

  • A simple landing page that looked clean and nice.
  • A to-do list app where you can add, delete, and mark tasks as done or completed.
  • Another page that worked functionally, even if the UI was not fancy.

The model actually works well for these tasks. The to-do app functions correctly, and you can interact with it. Julian noted that the UI might not be amazing out of the box, but you can train it or use a good skill to improve it.

You can use Laguna XS 2.1 in three main ways: as a local engine in Agent OS, through FreeClaud Code, or with Hermes agents. Each method is easy to set up and use.

05Results and Comparisons in the Episode

Julian compared Laguna XS 2.1 with other local models on benchmarks. On SWE Bench Verified, it performed close to Qwen 3.6, which is a popular local model. It also outperformed North Minicode, which was previously decent.

Against frontier models, it beat Haiku 4.5 on SWE Bench Pro and was far ahead of GPT OSS. However, Julian reminded viewers that local models are not the same as frontier models, so you should not expect them to beat the best cloud models.

He also mentioned that the model is about a year behind what you could create with Claude last year, but it is still a good alternative for simple builds.

06Tips, Limits, and Mistakes to Avoid

Here are some tips and warnings from the episode:

  • Do not trust benchmarks alone. Test the model yourself, as Julian does with Goldie Bench.
  • Lower your UI expectations or train the model with a good skill for better designs.
  • The model is not good at creating 3D models, so avoid that use case.
  • If you run local models, they can slow down your system. Using OpenRouter is a good alternative.
  • Make sure you have enough memory. Julian used 36GB, which worked well.

Julian advised to always test models in real scenarios. He created Goldie Bench to show real builds, not just company benchmarks.

07Who Should Use Laguna XS 2.1

Laguna XS 2.1 is for anyone who wants to run AI locally or for free. It is great for developers, hobbyists, and people who care about data privacy. You can use it offline, and your data stays on your machine.

It is also useful if you want to build mini apps or landing pages without spending money. The free API on OpenRouter makes it accessible even if you cannot run local models due to hardware limits.

Julian mentioned that you can plug it into your Agent OS and memory galaxy to improve its context and builds. This makes it a practical choice for those already using agentic workflows.

08How to Get Started with Laguna XS 2.1

To start using Laguna XS 2.1, visit Hugging Face to download the model. Alternatively, get a free API key from OpenRouter if you prefer not to run local.

Once you have it, plug it into your preferred setup, like FreeClaud Code or Hermes agents. You can create projects and see them in your workspace.

Julian also showed that you can create a Hermes agent profile with this model, allowing for agentic use. This gives you three options: local engine, FreeClaud Code, or Hermes.

For more resources, Julian recommends joining his community to see what others build with local models. He also plans to test it on his local leaderboards soon.