AIOS: 운영 체제에 AI 에이전트 도입하기

운영 체제에 AI를 도입하여 리소스 할당을 최적화하고, 컨텍스트 전환을 촉진하고, 풍부한 툴킷을 제공하고, 개인정보 보호 문제를 해결하는 AIOS 프로젝트에 대해 알아보세요. 성장하는 AI 커뮤니티에 참여하세요!

### AIOS: 대규모 언어 모델 에이전트를 통해 운영 체제에 AI 도입하기
* 🤖  AIOS는 운영 체제에 AI를 도입하여 작업을 자동화하고 컴퓨터에 AI 에이전트를 배포하는 프로젝트입니다.
* 🧠  Mixol 및 Quen과 같은 대형 Nish 모델을 운영 체제에 두뇌로 내장하여 OS에 영혼을 부여하고 리소스 할당을 최적화할 수 있습니다.
* 📦  이 아키텍처에는 대규모 언어 모델 관련 활동을 조정하는 에이전트 스케줄러, 컨텍스트 관리자, 메모리 관리자, 스토리지 관리자, 도구 관리자 및 액세스 관리자가 포함되어 있습니다.
* 🖥️  커널은 상담원이 서비스에 액세스할 수 있도록 통화 인터페이스를 통해 LM 시스템을 노출하며, AIOS의 설치 프로세스에는 저장소를 복제하고 필요한 패키지를 설정하고 Face ID를 허깅하는 작업이 포함됩니다.
* 📊  AIOS는 애플리케이션 계층, OS 계층(커널 계층), 하드웨어 계층으로 나뉘며, 개인화된 가상 코딩 도우미를 배포하는 등의 실제 사용 사례가 있습니다.

[Music] We have been seeing more and more frameworks being built to integrate AI into our operating systems. Today, we're going to be taking a look at AIOS, which is a project that is infusing AI into our own operating systems to automate, as well as deploying AI agents onto our computer. AIOS is a large language model agent operating system. It's something that embeds different large Nish models like Mixol, you have Quen, as well as many other LMS into our operating system as a brain of the OS. And it's something that enables the OS to have a soul. You can say it's a system that is designed to optimize resource allocation, facilitating context switch across different agents, enabling concurrent execution of agents, providing tool services, maintaining access control, and providing rich sets of toolkits for different large language model agent devs. Now, I want you guys to take a look at this figure right here, which is a figure that is motivating example of how an agent, a travel agent in this case, requires both large language model level

and an OS-level resource and functions to complete a task. So, in this example, it's utilizing both of these two factors to function on completing a different task. It involves organizing a trip based on the user's preference. So, we can see that the user is stating, I'm flying from San Francisco to New York for the for business next month. Please help me organize the trip. So, this query is then sent to the travel agent that is operating on the OS. So, whenever the agent interacts with the large language model service for tasks such as retrieving preferences, it makes decisions on tools as well as APIs that generate reviews as well as responses. It also interacts with the traditional OS services for tasks like accessing dis drives. You also have the ability to execute software. So, for each step that you can see over here, there's one step, two steps, three steps, etc. And in these steps, it involves a combination of a large language model reasoning as well as an OS-level action. So, it focuses on the preference retrieval. You have the

flight and well recommendation which is utilizing different things such as a tool API that is managed by the LM. You have an LM storage managed by the large language model. This storage, which is operated by the traditional OS, you have tool API software software text generation, and all of these steps are working alongside side with the agents that are deployed on our OS. Sorry for being repetitive, but this month we had insane Partnerships with big companies giving out subscriptions to AI tools completely for free. These are tools that will streamline your business's growth and improve your efficiency. Just being a patreon this past month you were given access to six paid subscriptions completely for free. Not only do you access these subscriptions, but you gain the ability for Consulting networking collaborating with the commun Community as well as with myself. You get access to daily AI news resources giveaways and so much more. If you're interested check out the patreon link in the description below to gain access to these benefits. Now, if you are to read the research paper further, it discusses the

challenges that are posed by increasing complexity and the quantity of Agents, such as the resource management of Agents, scheduling, as well as privacy concerns. So what this paper has done to address these challenges is that they proposed the AIOS architecture, which is depicted in this figure below. And this is something that we're going to take a look at as we go further, but in brief, it's an architecture that includes a large language model specific kernel, and it's designed to isolate large language model related tasks, as well as resources, from other OS functions. And this kernel is basically broken off into several different modules, and it's something that is really interesting, and it's something that you can actually deploy on your computer today. And it's something that we're going to be taking a look at as how you can get started so that you can deploy these different AI agents so that it can work on accomplishing various tools, such as an agent scheduler that prioritizes and schedules agent requests to optimize large Lage model utilization and such.

forth, I'm not going to be going over all of these because the intro is already way too long. But this is something that we're going to be taking a look at throughout today's video, so I hope you enjoy it. Stay tuned and let's get straight to it. If you would like to book a one-on-one with me where you can access my Consulting Services, where I can help you grow your business or basically give you a lot of different types of solutions with AI, definitely take a look at the calendar Link in the description below. Hey, what is up guys, welcome back to another YouTube video at the world of AI. In today's video, we're going to be taking a look at this project, which is aios large language model agents in our operating system. Now, we already mentioned how it basically functions, but let's actually take a deeper dive as to how this architecture operates. As I stated before, it functions through kernels that consist of several different modules. We talked about agent scheduler, but context manager is

something that supports snapshotting and restoring the immediate generation status. And this is for the large language model. It manages the context window as well. You have memory managers, which are providing short-term memory for agents that are interacting with different logs. You have store manager, storage manager, which is a persistent agent that interacts with logs in long-term storage. Tool manager, it is a manager that agents call for external API tools. And lastly, you have the access manager, which enforces privacy and access control policies between agents. These basically are different modules that work together to enhance the management and the coordination of large language model related activities within the operating system that you deploy AIOS within. So, this is how the overall architecture would work based off these kernels. Now we're going to go into a deeper dive as to how it basically functions. And this is where the kernel exposes an LM system through a call interface for agents to access these services. So, it basically functions through this whole architecture. It's kind of complicated, but if you are to get a deeper dive by reading through the

different layers which they specify you would get a better idea. Now before we go further into the video, I want to specify for the people who would be interested in installing this. It's kind of pretty easy. What you would want to do is you would want to clone this repository. So you would basically want to go into your command prompt. And there's a couple prerequisites that you would need which I should specify before. First, you need to make sure that you have Pip get as well as a functional hugging face Hub token ID. So once you have these things ready as well as the memory to host something like a gamma 2 uh two bilon parameter model, you should be able to host this quite easily. So what you're going to be doing is going into your command prompt. You're going to be cloning this repo. So what you would do is just clone this repo by copying it and clicking this green button which will give you the access to the link. You can scroll down go into your repo and type in get clone.

press enter. Once you have cloned this, you can then wait until it finishes cloning. Then you can go into this iOS folder by typing in CD iOS. Once you're in this directory, you can start installing the requirements. And they basically specify that make sure you have Python 3.9 or greater. And you need to install the required packages with this command. So you can simply just go in, paste this command in, and press enter. This will take a couple of seconds to a couple of minutes depending on your computer. But once it finishes doing that, in the meantime, you can work on finding your Hugging Face ID, which I'll showcase how you can do it. So to get your Hugging Face access ID, you can go on to Hugging Face. If you do not have an account, definitely create one. Once you have done that, click on this icon over here, which is your profile, and go to settings. Once you are in your settings, go to the access token key, create a new token. You want to name it iOS, and you can just

keep it whatever you want generate the token. Copy this copen ah token and then we can go proceed forward with installation. Now what you want to do next is then set your hugging face ID within a iOS. This is obviously after has finished installing the requirements. So what you would want to do is copy this command. Now if you are on Windows, you don't want to use the expert function. You want to use set as your command. And you would basically type in set hugging face ID everything normal just replaced your uh token in this uh section over here which is your read token. What once that is done you want to then do set hugging face and then your cach directory. And once you have basically set that you can run the main.py to start. And this is something that's really important. You can choose between these two LMS at the moment. You have gamma 2B as well as mistol 8x7 V. So what you would want to do in this case you want to make sure that you replace the

max GPU memory and the Evol device based off of your own requirements. And you can definitely find that on your own operating system or your own control panel, which will tell you the information on what's your max GPU. I currently do not have any requirements to run these models, so I'm not going to be testing it out. I don't want to find my fry my computer, but this is easy as that. You can then have the interface, which will pop up into your command prompt stating what you need to do next. It'll basically specify what sort of task you want to do with the LM as the brain of your operating system, and it can execute various tasks on your system. And do you can basically accomplish various things with the different modules that we mentioned before. And that's about it for getting started with aios. So let's actually take a look at the architecture once again because we have a couple components that we have left to take a look at. We have these aios layers. It's basically split up into

three layers: you have the application layer, you have the OS layer which is the kernel layer, and then you have the hardware layer. Firstly, with the application layer, it's where the agent applications like the travel agent, you have the coding agent, math agent are being deployed and developed. So what happens is that this system, the aios system, is providing the SDK at this layer which abstracts of the system calls are being used to simplify the whole development process for the agent. So it's working alongside with the aios SDK. It is then sending the related calls to the LM. If it's not needed for LM-based activities, it is then going to the OS kernel, which is then going to require the hardware functionalities. And this is where we go to the kernel layer, and it's basically consisting of the two main components which we mentioned before: the OS kernel and the LM kernel. The OS kernel handles the non-specific operations of an LM, whereas the LM kernel is dedicated for LM-specific tasks. And this segregation you can see is allowing for LM kernel to focus on the

critical LM-related activities. And these are the modules that we mentioned. They actually go a little bit more in-depth on each and every module, which gives you a better idea as to what you can do with these different agents when they're deployed on your OS. And lastly, we have the hardware layer. This is encompassing a physical component of the system, such as your CPU, GPU, memory, as well as many of the other components over here. Now, after reading a couple of sections of this report, I was just wondering, what are some practical use cases for AI OS? Like, why would you actually involve having the trouble to integrate AI into your operating system? Well, there's obviously going to be a lot of different use cases for different preferences, but one thing that would really benefit for me is that having it so that it would be used for coding. And this is where I would deploy different AI-powered large language models by infusing it into the operating system based off the framework that they have provided. And this would involve deploying a personalized virtual coding

assistant. This would help me for basic coding needs, filling out different prototypes, as well as helping me in various cases where I would need debugging. Now, this coding agent would be tailored for specific programming languages, and it would also help me with real-time support. And that was just one thing that I've really thought about, 'cause it could help me with auto-completion, code snippets. It could help me with error detection, as well as syntax highlighting. And this was one thing that I was just thinking that it could assist me in. And I would just wonder what other people would do with this, and I really always, whenever I make videos on this, I know these tools are really helpful, but in practical use cases, I don't really try each and every one of them out. But I would really be intrigued to see what my viewers are building and doing with these different tool sets. So, if you're interested, you can even pitch me an email, showcase what you've built with it, and that I can possibly even make a website.

future where I compile all the tool sets that I showcase, and all the US cases that people have developed. And this is like the community aspect that I really want to develop with this Channel. I want to have an interactive AI community that would be basically pitching out different ideas and different tools. I already do that with the Discord, but I would be interested in having a more open-source like Community where everyone can contribute to that. So if you're interested in that, definitely send me an email on what sort of things that you would build with all the tools that I cover in my like videos. And I'd be interested to see what you guys have been building. And obviously if it's super interesting, I would definitely make a video on it. But I know I'm rambling too much now. That's basically it for today's video. I hope you enjoyed it, guys. This is definitely a great tool that I personally see some sort of use case out of for deploying on your local computer. It's basically giving a brain to your OS. And this is something

that we have mentioned previously in this video. So, I hope you enjoyed this video, guys. If you haven't, definitely check out the Patreon page. This is a great, great way for you to stay up to date with the latest AI news, as well as accessing different subscriptions. We have a growing community, and it's just growing like tremendously over the past few months. And I truly recommend that you would love it. And if you are to join it, you would definitely benefit from. Make sure you check out the Twitter page if you haven't. This is a great way for you to stay up to date with the latest AI news. And lastly, make sure you guys subscribe, turn on the notification Bell, like this video, check out our previous videos. There's a lot of content that you would definitely benefit from. So with that thought, guys, have an amazing day, spread positivity, and I'll see you guys fairly shortly. Peace out, fell

AIOS: LLM Agent Operating System! Create Software, Automate Tasks, and More!

World of Ai is here to push the creative use of Artificial Intelligence (AI) applications for simple day-to-day tasks!

Here at WOA, you will find tips, tricks, and guides using AI applications to help you out in your day-to-day life. Going to strive to make your life easier! I will show you how and what you need to do to help your life out! I will strive to show you the wonders of AI and if that interests you, then make sure you subscribe to the channel now!


WorldofAI

AIOS는 대규모 언어 모델을 운영 체제에 통합하여 작업을 자동화하고 AI 에이전트를 배포하는 프로젝트입니다.

리소스 할당을 최적화하고 에이전트의 동시 실행을 지원하며 대규모 언어 모델 에이전트 개발을 위한 툴킷을 제공합니다.

이 시스템은 에이전트와 운영 체제 간의 상호 작용을 허용하여 대규모 언어 모델 추론과 OS 수준의 작업을 결합하여 작업을 완료할 수 있습니다.

또한 이 프로젝트는 컨설팅, 네트워킹 및 AI 리소스를 포함한 무료 AI 도구 구독과 Patreon 프로그램을 통한 혜택도 제공합니다.

AIOS 아키텍처: 대규모 언어 모델 에이전트 관리

AIOS 아키텍처는 상담원 리소스 관리, 스케줄링 및 개인정보 보호 문제를 해결합니다.

여기에는 작업, 리소스 격리용 모듈이 포함된 특정 커널이 포함되어 있으며 배포에 사용할 수 있습니다.

이 아키텍처는 에이전트 스케줄러, 컨텍스트 관리자, 메모리 관리자, 스토리지 관리자, 도구 관리자 및 액세스 관리자로 구성된 커널을 통해 운영 체제 내에서 언어 모델 활동의 관리와 조정을 강화합니다.

커널은 에이전트가 서비스에 액세스할 수 있는 호출 인터페이스를 통해 LM 시스템을 노출하며, 복잡하지만 포괄적인 아키텍처로 작동합니다.

허깅 페이스 모델 호스팅 설정하기: 단계별 가이드

허깅 페이스 모델을 호스팅하려면 리포지토리를 복제하고 Pip과 작동하는 허깅 페이스 허브 토큰 ID가 있는지 확인해야 합니다.

복제한 후 iOS 폴더로 이동하여 특정 명령을 사용하여 필요한 패키지를 설치합니다.

그런 다음 프로필 설정에서 허깅 페이스 액세스 ID를 얻고 새 토큰을 생성한 다음 토큰을 복사합니다.

마지막으로, iOS 내에서 명령을 사용하여 허깅 페이스 ID를 설정하고, 허깅 페이스 캐시 디렉토리를 설정한 다음, main.

py를 실행하여 사용 가능한 LMS 옵션 중에서 선택할 수 있도록 합니다.

AIOS(AI 운영 체제) 아키텍처 이해

AIOS 아키텍처는 애플리케이션 계층, OS 계층, 하드웨어 계층의 세 가지 계층으로 구성됩니다.

애플리케이션 계층은 에이전트 애플리케이션을 배포하고 AIOS SDK를 활용하여 개발 프로세스를 간소화합니다.

OS 계층은 LM의 비특정 작업을 처리하며, OS 계층의 LM 커널은 LM 전용 작업 전용입니다.

하드웨어 계층은 시스템의 하드웨어 기능이 필요한 곳으로, LM 기반 활동에 필요하지 않은 경우 OS 커널과 상호 작용합니다.

운영 체제에 AI를 통합하는 실제 사용 사례

코딩에 AI를 통합하여 기본적인 코딩 요구사항, 프로토타이핑 및 디버깅을 위한 맞춤형 가상 코딩 도우미를 제공하고 특정 프로그래밍 언어에 맞게 맞춤화하며 실시간 지원을 제공할 수 있습니다.

AI 기반 시스템은 자동 완성, 코드 스니펫, 오류 감지, 구문 강조 표시 등의 기능을 지원합니다.

사용자는 잠재적인 기능 쇼케이스를 위해 자신의 경험과 창작물을 AI 통합 기능을 통해 공유하도록 권장됩니다.

AI 도구 개발을 위한 커뮤니티 참여 모색

이 연사는 사람들이 자신의 아이디어와 도구를 기여할 수 있는 오픈소스 대화형 AI 커뮤니티를 만들고자 합니다.

시청자가 소개된 도구를 사용하여 구축한 프로젝트를 이메일로 보내면 동영상에 소개될 수 있도록 장려하고 있습니다.

또한 이 발표자는 Patreon 페이지를 홍보하고 시청자들에게 최신 AI 뉴스와 구독을 위해 성장하는 커뮤니티에 가입하도록 독려합니다.