What is a local agentic OS and how it works
A local agentic OS is a complete AI agent system that runs directly on your own computer, without needing the cloud. In this episode, Julian Goldie shows a new local AI model called LFM 2.5 that works inside his agent OS, called Hermes agent. This setup lets you search the web, create skills, and remember information, all offline.
LFM 2.5 is a tiny AI model with only 2.6 billion parameters, but it was trained on 34 trillion tokens of data. It has a 128,000-token memory window, so it can handle a lot of information at once. Because it's small, it runs fast even on a normal laptop with just 8 gigabytes of memory. The best part is that all your data stays on your device, so nothing you type ever leaves your machine.
Main features of LFM 2.5 in the agent OS
LFM 2.5 has several standout features that make it great for local use. Here's what Julian highlighted in the episode:
- Offline operation: Works without Wi-Fi, so you can use it on a plane or in a coffee shop.
- Built-in tool use: It can search the web, create skills, and complete tasks step by step, all on its own.
- Privacy by default: Since everything runs locally, your data never goes to the cloud.
- Speed: It responded quickly and didn't slow down Julian's system, unlike other local models he tried.
- Training inside an agent system: It was trained inside Hermes agent itself, so it already knows how to act like a real assistant and use tools properly.
- Reusable skills: It can learn a new skill from a guide and save it for later use.
These features make it a powerful tool for anyone who wants AI automation without the usual cloud dependencies.
How to set up your local agentic OS
Setting up your own local agentic OS doesn't require technical skills or building anything from scratch. Julian has packaged the entire system into a ready-to-install file. Here's how to get started:
- Get the agent OS file from the link in the description – it's inside the AI profit boardroom.
- Download the file to your computer – all you need is a normal machine with at least 8GB of memory.
- Install the agent OS on your device – it's a simple setup, no complicated steps.
- Connect LFM 2.5 to your agent system – it plugs right in, as Julian showed.
- Start with a small task, like asking it to search the web or look through your notes.
Julian also mentions that you can join the AI success lab, his free AI community, for full setup notes and more free tools. The link is in the comments and description.
How to use the local agentic OS with real examples
In the episode, Julian demonstrates several practical uses for the local agentic OS. One example is asking it to check his local notes and give content ideas for his AI community. It searched through his saved notes, found patterns, and returned a list of ideas he could use for videos and posts – all in seconds.
Another example involves skill building. Julian gave the model a guide and asked it to learn from it. The model searched the web on its own, read the guide, and saved a working skill for future use. This is something that used to require manual setup.
He also shows coding inside the agent OS. He asked it to build a simple landing page for the AI profit boardroom, demonstrating how AI automation helps get leads and customers. The model built the page right there, and Julian could preview it instantly.
For content creation, you can ask it to check your memory and give keyword ideas based on your existing notes. With a simple request like "check my memory and give me keyword ideas," it comes back with specific tool names, setups, and comparisons – without a long prompt.
Results and comparisons from testing the local agentic OS
Julian tested LFM 2.5 extensively over a few days. In early testing, it performed better than some other well-known small models, while also running faster. The combination of speed and smart tool use is rare for a model this small.
When he ran the local agentic OS, it called tools without any problems. It searched the web using a built-in tool, and even learned and saved a new skill on its own. All of this happened locally, with no data sent to the cloud.
Julian notes that his whole system usually slows down when running local models, but this one didn't slow anything down. It responded quickly, called tools correctly, and never got stuck. This is a significant improvement over other local models he's tried.
Tips, limits, and mistakes to avoid with the local agentic OS
To get the most out of your local agentic OS, keep these tips in mind. First, don't use LFM 2.5 for everything. Instead, use a bigger, smarter model as the brain for complex decisions, and let LFM 2.5 handle repetitive tasks like web searches, note organization, and simple skill creation. This saves your best model for hard thinking.
Second, start small. Let the model search the web, look through your notes, or build one simple skill first. Once you see how fast and reliable it is, you can hand it more busy work.
One limitation is that it's a small model, so it may not be as powerful for complex reasoning as larger frontier models. But for everyday tasks, it works great. Avoid the mistake of expecting it to do everything – use it where it shines, which is speed and tool use.
Who should use a local agentic OS and why it matters
This local agentic OS is ideal for content creators, business owners, and community managers who want to automate tasks without sending data to the cloud. Julian emphasizes that it's perfect for anyone who cares about privacy and wants to use AI anywhere, even without Wi-Fi.
It matters because it shifts how you use AI. In the old way, every message went through the cloud, requiring a lockdown setup and watching usage limits. Now, with a local setup, you get speed, privacy, and unlimited use – all for free. You don't need a fancy computer or technical skills. This makes powerful AI practical for regular people running real businesses.
As Julian says, it's the whole point of AI: making it practical, not just impressive. If you're just getting started with AI agents, this is a great place to begin.
