An open-source UI-focused agent framework that translates natural language user requests into actionable operations on Windows OS.

Okay so summarize the article on screen then write a summary with a dystopian Outlook in my docks. 

Okay then just download the summary as a PDF. 

Okay so let me run this I'll be back in a few minutes. 

Okay let's see here is the task complete. 

Yeah that looks quite good do we have the PDF? 

Yes we do. 

Okay perfect. 

Okay so what you just saw here was me running Microsoft UFO. 

This kind of aligns with what I think can be the future of computers. 

So let's take a deeper look at what the AI PC could be. 

Okay so here is kind of how I thought about the AI PC. 

So of course we will need very powerful hardware, high-end GPU, good CPU, a lot of memory right to load all the models. 

And this PC will have like built-in AI software. 

So LLMs, image generators, video generators and all that kind of nice stuff. 

But the key here is that we can run this AI software locally on your computer. 

So we will not be bound to the cloud because the cloud introduces a lot of high latency. 

Right we don't want that. 

Uh and of course it's going to be like I think it's going to pay off running this locally in terms of prices when if the GPU prices.

Come down and stuff and the idea behind this is just instead of operating the computer yourself you can just give instructions like we did in the intro. These instructions will be executed by the built-in soft AI software, right? And all you have to do is just sit back and monitor this execution, maybe confirm a few high-end critical operations. 

Maybe you can run a lot of parallel VMS and stuff and yeah that is kind of the basic IDE behind the AI PC, at least my ID of it. Before we take a deeper look at the Microsoft UFO, uh, I just wanted to show you this article here and this PC. So this is kind of the idea I had. Someone took Nvidia's faster CPU ever and built an absurdly fast desktop PC with no name. 

It cannot play games but it comes with 576 gab of RAM and starts at $43,000. So this is kind of the idea. So this is the world's first Nvidia GH2 200 Grace Hopper desktop super computer. So this is kind of everything I was talking about. So if we go take a look at this high-end computer, right, you can see all the specs it has. 

So if you see what are the benefits, it has enough memory to run the biggest LLMs.

Currently available locally that is kind of crazy right. The lowest possible latency very power efficient easy to transport very quiet. So you can see this is perfect for kind of on the edge and yeah that is kind of what we're looking for in this aipc right. And it's also cheaper and much faster than Cloud providers. So this is exactly what I was talking about and this is what I kind of expect that we can see more of in the future right when the prices come down.

Of course if you we take a quick look at the technical details here these are just insane. So they are available Aero liquid cool an Nvidia g200 gray Hopper super chip 72 core Nvidia gra uh CPU an Nvidia h00 tensor core GPU 480 DDR I think memory 576 GB of Fast Access Memory. So yeah this will be pretty crazy it runs on two times uh 2,000 Watts PS psus and yeah these stats are just off the sh. This is kind of exactly what I was thinking about like so I think this was perfect timing for me because I haven't seen this before and it kind of gives you the idea of what these AI pieces could be in the future right. But of course the hardware is only half the part.

Of course you need the software to utilize these uh good hardware stats. 

And today we're going to take a look at Microsoft UFO. So this is a UI focused agent for Windows OS interaction. This is kind of what you saw in my intro. UFO is a UI focus dual agent framework to fulfill user requests on Windows OS. So uh I think we're going to test it out some more. I'm just going to show you quickly how you can install this and try it out for yourself. 

So we're just going to go through quickly these steps here. So of course this is only for Windows now, but let's do it anyway. So let's copy this. So we're just going to go into our folder. So let's clone the repo. Okay that was done. Now let's go into directory right and let's do pip install. I will probably have all of this installed now, so let this run. 

But the next step is uh we are actually using OpenAI here, so you will need an OpenAI key for this. So we're just going to go into our config file and kind of set up everything we need here. So here we are inside our config do yaml. So we're going to have to set our end point. So we can just go back to the GitHub here and just.

Copy this endpoint here so it's just the chat completions endpoint point. And let's go back here and then you have to fill in your API key. And for this, we are using the wish model. So this is you can just keep this as is. So the only thing you have to fill in is kind of the endo and your API key and you should be ready to go.

And to start this from the Windows terminal or the command line, we just have to run this python DM UFO and kind of task and just give it uh a name. So let's try this. Okay, so let's run it now. You can see kind of please enter your request to be completed. 

So what I thought we can do, we can just uh try to grab some stats from this website and we have kind of our email here. So we're going to try to send an email. I think you have to confirm this but we will just see what happens. 

So let's write our instructions. Okay, so let's start off pretty simple. So from the website on the screen, summarize the technical details to send it as an email to Chris at allabout AI. 

Okay, so let's fire this up. So we're just going to let this run and take a look at the outcome after. And just a warning, this is very expensive.

To run so yeah because it's using a lot of data I think uh but let's look at that after. So let this run and see where this takes [Music] us. 

Okay so I just wanted to bring you back for this because now input required. So we will have to confirm if we want to send this email. Everything looks great so far. We have the technical details so I'm just going to press uh y here right and see what happens. 

Okay so we have task completed there we got our email perfect. But now here comes the bad part. So if I exit now right we spent $1 on this that's way too expensive. So this is kind of what I was thinking about with the AI PC. This cost would be like we didn't have to pay anything because we did all of this like locally on edge and this is just crazy expensive. 

So I would not recommend you doing this just take a look at what I do now and maybe try again in a year or two because it's crazy expensive uh but we got our email perfect with all the technical detail. It's a bit weird but task is completed and if you go scroll up here look at all the planning and how to do to complete all this it's crazy and I think it did a very.

Good job. So we kind of number one switch to email draft and it kind of uses some kind of if you look at the buttons here. So you can see click input button left double false on the send item. So it's using the wish to kind of locate all the buttons and stuff. I think instead of using like the Dom or I think it's just using GPT for vision. 

So let's try a few more examples and spend some more money. Okay, so let's give it a more challenging task. So ask chat GPD create a python code to generate Fibonacci numbers. Copy the code, paste the code into VC. Don't write it, run the code in VSC. So let's see how this goes. 

Okay, so that didn't work. Uh, it wasn't too far away. I think it kind of did all the steps correctly. So let's see how much this cost me. $1. Oh God, that's expensive. Uh, it didn't, it didn't paste in the code. So I don't know how to do that. 

Let's just for fun, let's paste it in, let's run it. Uh, I don't even think it works. Everything, every step was kind of good all the way up to I couldn't figure out how to paste it in. That was kind of the fail point. But other than that, I think he did a pretty good job at least. That is kind.

Of what I had today h so I'm quite curious to think. Do you think this aipc will ever be a thing? Leave a comment below so I can read uh what you guys think. 

Uh but I had a lot of fun with this and it kind of shows that some software is kind of trying to be created. But I think it needs to be much better than it is now. 

Uh follow the link in the description below if you want to try out the Microsoft UFO but be very careful it's very expensive. 

Uh other than that share AI Agent Store with colleagues if you thought it was interesting and I see you again on Wednesday.

Note, that there may be slight differences in text and video. Watch video for the original content.

UFO Video