About this video
Docker is a prison for your AI productivity. Last week we set up PewDiePie’s Odysseus using Docker, but today we are breaking it out of the sandbox to give it real power over your machine. Key Takeaways: - Why Docker limits the agentic capabilities of Odysseus. - Hardware recommendations including Mac Mini and RAM considerations. - How to install Odysseus locally on macOS, Linux, or Windows. - Setting up local models like Gemma 12B using MLX for peak performance. - A live demo of the AI agent interacting with local system files. - How to expose the interface to your phone via a trusted LAN or VPN.
Why Relying On Cloud AI Is A Strategic Mistake For Your Data Security
Most users are currently trapped in a cycle of dependency on cloud providers like OpenAI and Anthropic. Whilst these services are convenient, they act as a black box for your sensitive information. If you want true privacy and the ability to let an AI actually interact with your computer, you need to stop using Docker and start running Odysseus locally.
The Limitation of the Sandbox
Docker is fantastic for encapsulation, but it acts as a prison for agentic tasks. When you run Odysseus in a container, it cannot easily 'see' or interact with the files on your host machine. This defeats the purpose of having an agent that can organise your downloads or manage your local environment.
Hardware Requirements
You do not need a supercomputer to get started. If you are tapping into existing API subscriptions, a Raspberry Pi or an old laptop will suffice. However, if you want the gold standard of privacy, local models are the answer. A Mac Mini with plenty of unified memory is the perfect companion for this task. The more RAM you have, the larger the models you can run.
Step-by-Step Local Installation
The process involves using the Terminal to clone the repository and run the setup scripts directly. This allows the application to run 'close to the metal', granting it the permissions necessary to perform complex tasks.
- Clone the Repository: Use the git clone command to pull the latest Odysseus code.
- Navigate and Initialise: Change directory into the folder and run the start script specifically for your operating system.
- Set Up Access: Define your admin credentials and ensure your LAN settings are configured if you wish to access it from other devices.
Choosing Your Models
For Mac users, MLX models are far superior to GGUF variants because they are optimised for Apple Silicon. By using tools like OMLX or LM Studio, you can download models such as Gemma 12B and link them directly to Odysseus via a local API.
The Power of Agentic Mode
Once Odysseus has access to your machine, you can ask it to perform tasks that ChatGPT simply cannot do. It can scan your folders, identify large files, and make recommendations based on your actual data. This is the future of personal computing: a private, local agent that understands your specific environment without ever sending a byte of data to a remote server.
Stop playing in the sandbox and start building a real local powerhouse today.
Transcript▾
Last week I did a video on PewDiePie's Odysseus and I kind of left you hanging because we set it up on Docker which produced a lot of limitations with regards to performing agentic tasks across your machine. This is what Docker is for: encapsulation. It protects things. To get the true benefit of something from Odysseus if you're using it outside of its very capable chat capabilities, you're going to want to run it on your local machine and that is exactly what we're going to be doing today.
Now, I would fully recommend doing this on a separate computer. A Mac Mini is traditionally the one to use this on especially if you're using local models. You can run quite good local models on a Mac Mini. However, with all of this stuff as I've explained in many videos, the more RAM you have the better. However, if you're not running local models and are tapping into your ChatGPT subscription, you can literally run this on a very old PC that you've got in the closet or a Raspberry Pi. It doesn't need anything powerful to run Odysseus. That doesn't change the installation process which we're going to run through today and I'm going to talk through a lot of the local model considerations that you might have.
All that to be said, if you want to set up Odysseus to run on your local machine, then this is exactly the tutorial for you. So, Odysseus. If we go to get started because it's a lot easier to do it on GitHub here. If we scroll down to the getting started, this is the one we ran last time which is getting Docker. Ignore that. Depending on whether you're on Apple Silicon, Linux, or indeed Windows, these are going to be the instructions you're going to run.
So, if we copy this first line here and I'm aware a lot of people in the audience probably don't have the foggiest idea about all of this technical stuff. Unfortunately, this is the nature of doing this stuff. You're starting to get into a very technical world. I'll do my best to explain things. Long story short, we're going to open up terminal here. My favourite terminal here is Warp. I'll leave links to this down below. This allows you to run instructions on your machine. Pretty much every application is under the hood running these sorts of commands. So, it's really bare-bones, as close to the metal as you call it.
Macs just have a terminal. If you just open up your terminal app, these are the commands you're going to run. So, all we're going to do is CD, which is change directory, and I would recommend getting your finder and dragging in a folder. I am going to do my sites one here, just dragging that into there, and this will make it so that you're inside of this folder. Then, I'm going to paste this, and what this is going to do is clone or pull down the code that PewDiePie has written. And now, we can CD into the folder that it's created.
The instructions even say to do that. Now, we're going to go to start macOS, which should launch it up at this URL. To expose it on your phone over trusted LAN or VPN, such as Tailscale, I would probably recommend that. We need to run this. So, actually, we're going to run that one, and this will mean that you can use it on your phone. You can use it on many other computers or just keep it on your local machine.
While it's doing that, let's check the instructions here. It says here you can build a clickable app wrapper, which means we don't have to run this in terminal every time, but it will create a little application. We're being nerds at the moment, so we're going to stay in the terminal.
It's asking us for our username. We're going to leave it as admin and give it a password. We have it up and running. Admin. Password. And we are in.
Quick run through. This won't be a deep dive on Odysseus, but we can run a chat mode, which is exactly as you expect. Agent mode is actually acting on your machine and this is in my opinion one of the most powerful things you can do. You can add your email, add tools, calendar, and stuff like that.
Let's get some models set up. You can use OpenAI, Anthropic, or DeepSeek. One of the most tempting things with Odysseus is the local models. I have Ollama installed, but I'm on a Mac. What you want to look into is something like LM Studio or OMLX, which allow you to download MLX models. I have already downloaded the 12 billion parameter Gemma model.
If we go to agent mode, I can now say, "Look in my downloads folder and list out the largest files." You see it's running some tools. It's doing its thing. If we go to activity monitor, you can see my GPU is getting maxed out. We're pushing 74 gig of RAM here. So, for those that are interested, if you've got a 96 gig RAM laptop, you'll be able to push this Gemma model pretty well.
I hope this gives you a clearer idea of Odysseus and what you can do from an agentic standpoint. We have it running, it has access to our machine, and we can chat with a local model. This is completely private. Let me know down below if you want me to go deeper. Until next time, keep on vibing.