Magentic‑One + Ollama – Human‑Centered AI, Manus‑Killer!

Brief Summary

This video introduces Magentic UI, an experimental, human-centered web agent by Microsoft, powered by autogen, designed to give users more control over AI-driven web operations. It draws a comparison with Manus, a similar tool from a Chinese company, highlighting Magentic UI's focus on keeping humans in the loop for decision-making. The video further explains the architecture of Magentic UI, which includes an orchestrator, web server, coder, and file server, and demonstrates how to install and run it locally using Olama, showcasing its capabilities through practical examples like making restaurant reservations and summarizing repositories.

  • Magentic UI is a human-centered web agent that keeps users in control.
  • It is powered by autogen and can be run locally using Olama.
  • The agent can perform tasks like making reservations and summarizing repositories, with a live view of its actions.

Introduction to Magentic UI

Magentic UI is introduced as an experimental, human-centered web agent from Microsoft, designed to give users greater control compared to fully automated agents. Unlike systems where the AI performs tasks independently, Magentic UI keeps the human in the loop throughout the process. The agent is not fully available like Manus, a Chinese company's similar tool. Manus can answer questions and perform tasks such as booking restaurant reservations by creating a plan and executing it, showing the process on a virtual computer screen. Magentic UI is presented as Microsoft's answer to Manus, powered by autogen, which is a framework for building AI agents capable of orchestrating and using various tools to achieve goals.

Magentic UI Architecture and Features

The architecture of Magentic UI includes an orchestrator, web server, coder, and file server. The orchestrator manages web operations, coding, and file control in response to user queries. The system has been evaluated with simulated users, showing accuracy levels close to human performance. The video transitions to demonstrating how to run Magentic UI locally using Olama, which avoids the token costs associated with OpenAI.

Installation and Setup with Olama

To install Magentic UI, you need to create a virtual environment, install the Magentic UI package, and then run it on a specified port. The video shows the presenter's code, which is the same code used to create agentic multi-agent systems using autogen, including an orchestrator. The presenter demonstrates creating a virtual environment, installing Magentic UI with the Olama option, and running the application. If using OpenAI, an environment file with the OpenAI key is required; otherwise, it can be removed to run with local Olama, but some configurations are necessary.

Running Magentic UI Locally

Running Magentic UI requires Docker to be installed, as it operates within a container. The presenter starts Olama and then runs Magentic UI on port 8081. The system checks for Docker, a VNC browser image, and a Python image before launching the web application. After navigating to the specified URL, the Magentic UI interface appears, and a new session is created. To configure it to work with Olama, the model configuration settings need to be adjusted to select Olama and specify the desired model.

Interacting with Magentic UI

After selecting the Olama model, users can interact with Magentic UI, which uses the local large language model to process requests. The presenter asks the UI to create a markdown file summarizing the Microsoft autogen repository. A live view on the right side of the screen shows the operations being performed, similar to the Manus computer. The presenter then asks the UI to show code, specifically in C++, to see the plan that the UI will execute.

Human-Centered Control and Task Execution

The presenter asks Magentic UI to book a reservation at a famous Mexican restaurant in Auckland, New Zealand, specifying Mexican cuisine. The UI generates a plan to search for restaurants on Bing, considering reviews, cuisine quality, and customer ratings, and then navigate to the chosen restaurant's website to fill out reservation forms. The presenter adds a step to book the table by 8:00 p.m. on Friday, demonstrating the human-centered aspect of the agent, where users have more control over the plan before execution. After accepting the plan, Magentic UI starts performing the operations, which may be slower due to running locally.

Refining the Plan and Completing the Task

The presenter is prompted to provide a preference for the type of Mexican restaurant, and they specify "traditional." The plan is updated based on this input, and after accepting the revised plan, Magentic UI continues executing the task. The UI finds a traditional restaurant, Mexican mexical fresh new market, on TripAdvisor and provides details such as reviews and ratings. The next step involves navigating to the restaurant's official website for booking. The presenter then asks the UI to book the reservation from the mentioned site, which creates a new plan and initiates a new operation.

Final Thoughts and Examples

The presenter reflects on Magentic UI's ability to perform tasks and control the system, similar to Manus. They mention previous successful plans, such as making reservations, finding Microsoft research publications, and creating new logins for websites. The presenter concludes by highlighting Magentic UI as a cool, human-centered web agent built on autogen, giving users more access and control over operations.

Summarize any YouTube video in seconds

Get summaries like this one for any video, in your language, with the free Summ app.

Download on the App Store
Get it on Google Play

Transcript

Full transcript of “Magentic‑One + Ollama – Human‑Centered AI, Manus‑Killer!”. Click a timestamp to open the video at that moment. Need another video? Use the YouTube transcript generator.

0:00 Hi guys, welcome back. And today in this we'll be talking about yet another AI agent called as magentic UI. This is an experimental human- centered web agent which is going to be giving users or the humans the more ability to control the agent where the agent itself is not going to be doing everything for you. You as a human will be in loop to perform all these operation. That is what is this magentic UI of Microsoft is going to do. as it names it is an experimental human- centered web agent which means it is still not available uh in an full-blown version like Manus if you really heard about it before a Chinese company which does exactly the same kind of operation so you can go and ask some questions over here this is going to do things for you so for example if you're going to ask like uh can you uh book a um reservation for me

0:47 in the famous uh restaurant uh in uh Oakland uh NZ Z with Mexican cuisine. You see that once you ask the question to the manus over here, it is going to go and create a plan for you and the plan is going to be uh started building over here in a small computer as you can go and uh use this manus computer use uh you just go and uh maybe you can even open this uh in the uh in another window over here. Look at that. So it's going to show the computer uh which is going to be performing this operation. So uh at the moment the manus is doing things for you. That's the reason why manus computer is inactive. But the moment it's going to start working or uh perform any of the operation it will

1:33 show you the screen and it will show you how these operations are going to happen over here. Look at that. So this is going to go and search uh and it's going to perform this operation. This is what Manus does and this is exactly what uh Microsoft has built uh as an answer for the Chinese company manus over here uh in uh over here called as magentic UI. This magentic UI is actually powered by autogen. Again, autogen is something that we have already talked in our earlier video as well where we talked how we can build our own aenetic multi- aent systems uh and also how the aenetic multi- aent systems can have the ability to do the orchestrations and create or call multiple different tools to perform or achieve a single goal that you actually gave it to. That is exactly what this u autogen is going to do. It's

2:19 pretty much like lang chain or llama index or Google's ADK which do which does the building of the AI agents uh and rack systems and things. That's exactly what the autogen is all about as well. Uh so that is how this uh magentic UI is actually powered off. Uh and as I told you this magentic UI over here is going to also have like a live view over here as you can see where it's going to show you what's exactly going on while you ask any question to this particular agent. It is going to go and open a web browser and it's going to do all these operation for you. Uh and then it is going to uh create a plan. It's also going to show you the plan like how the execution is going to happen and then it is going to perform those operation for you on your behalf. And again and all these operation you as a human are going to be in loop. That is the major benefit

3:06 of having the uh Microsoft's uh magentic uh one UI over here. So you as a human are more controlled uh in this entire operation. Uh as you can see in all these videos that is what exactly what uh we are going to be discussing in this particular video and I will show you how we can make use of this um magentic uh one or magentic UI to perform this operation. As I told you again this architecture of the magentic UI is going to have an orchestrator web server coder and file server. So the orchestrator is the one which is going to perform all these operation for you. uh and uh it is going to control the uh the web operation, the coding operation and also file control everything for you while you ask any of the question to the uh the magentic one over here and also you

3:52 can see that they have got some evaluation result with the simulated user and you can see how better the uh the accuracy of the magentic UI uh and how pretty close they are with the humans itself. This is quite scary but but yeah that's what they have told us over here. I quickly show you how actually I uh run this particular magentic uh UI along with our local large language model which is using the OAMA instead of actually using the open AI which is going to take lot of tokens to perform this operation. You can actually run all of these uh in the uh in your local Olama itself. So if I go and click this particular GitHub over here, it's going to go take me to this particular URL. And if I just go scroll down, you can see that you need to

4:37 create a virtual environment and then you need to install the magent UI. Uh and then you can run over here on this particular port number or you can uh yeah, you can also uh run based on the Azure or Olama over here as they have mentioned. So I have already did that all installation part over here. uh so that's the reason why I have got this particular uh window opened and actually this is the same code which I was showing you how we can use the autogen to create the uh agentic multi- aent system as you can see over here it also has got an orchestrator which is nothing but the um the planner uh which is going to do all of these operation by invoking multiple different agents. This is exactly the same thing which is done by um by this guy as well the magentic UI

5:25 as well to perform these operation for you over here. All right, same thing same concept no big difference between both of them. Uh so one more time I can quickly show you what I really did while I was actually working with this particular magentic UI. I have actually created the um uh the the virtual environment over here and also I installed this magentic UI something like this and then uh I installed the uh pip install magentic UI lama because that is what I really require. So this is one thing which I wanted to quickly show you how it it needs to be done. So you just have to do uh colon and then uh magentic UI something like that because that is what you should be doing over here. So over here and if you hit enter this

6:10 is going to install the magentic UI Olama for you and then you just have to run this guy over here right and most importantly while you run this uh magentic UI over here uh if you're running with the um with the uh open AI uh you can create an environment or env file with the open AI key and and do this over there but if I are not going to use the open AI you can just remove that env file and then you can just run the uh magenta UI - port uh 8081. It's going to run in the local OAM for you. But you need to make some configurations and things. Uh I'll first open my Olama running within my machine. And I'm going to run the magentic UI in the port uh 8081. That's going to uh run things. But

6:58 look at that. While I start running this particular magentic UI, it is also going to look for a docker to be up and running. So you need to make sure that you also have docker installed in your machine. Uh if not it is not going to run because this is going to run on an container which is going to act like a machine which is where the executions are really happening over there. You see that I have actually got the uh containers which has got this uh thrusting viskoff which actually is basically from the uh magentic one which is going to be instantiated. See the moment I start this and while I start running the uh the magentic UI you see that the docker is going to be checked it's going to also check for the VNC browser image uh and then it's going to check for the python image and launching

7:45 the web application and now it is running things for you over here uh in this particular URL. So I'm going to go to this particular URL uh and look at that the manus is currently doing things for me uh which is great. I'm not going to worry about that. And I'm going to go open this particular URL. See, I tried some things over there. That's the reason why it's showing me all of these, which is amazing. I'm just going to leave that as it is. And I'm going to create a new session. And you can see that the the magentic UI comes in over here. In order to work with the uh the Olama, you just have to go to the settings and go to the advanced um sorry, the model configurations. Uh and over here, make sure that you choose Olama there. That is very important. You need to choose the Olama in order for

8:31 you to run uh this uh this guy with the Olama. If you're going to be using OpenAI which is the default, it is going to be taken from here and make sure that the env file on the particular directory exist. But because I'm going to use the Olama local, I'm going to choose the Olama over here. And I need to choose the model. Uh so my Olama will have different model from the one which is shown on that particular uh UI. So I need to go uh see the list of Olama that I have got and I have got these many models. Uh and let's say I'm going to take this uh Llama 3.2 latest uh model and I'm going to go paste it over here. Uh you don't have to save it to be honest. Uh and then you can just close it. It is by default going to be chosen from there. See that this is the thing

9:17 is going to be taken. Uh and now you can start interacting it. And this guy is going to work with the power of your local large language model. Uh over there over here I'm just going to probably ask can you make a markdown file with the Python that summarizes the Microsoft autogen repository. Let's see if that really going to happen or not within my local machine. Uh and look at that. Now the moment I do this immediately on the right hand side there is going to be a live view which is actually going to show me uh what exactly is going to happen on this particular screen which is pretty much like the manus computer that we have got which is going to do the same kind of operation uh that we have over here as well in the magentic uh UI and we can probably say uh we can now go and ask

10:04 some questions if we really have any over there. So in this particular UI um now I can as a user I can go and ask the questions over here uh saying uh the code uh show me code especially uh in the uh uh C++ C++ and if I hit enter uh if everything goes fine it is going to go and uh the magentic UI is going to go and show me the plan that it is going to go and execute uh on the right hand side on the particular screen to perform this operation and if we like the particular plan we can then execute the plan from there. So just waiting for this to happen and because it is running within my local OAMA this process takes a bit

10:52 of a time or maybe more time because it depends on the complexity of the question that we are asking in these things will change but if you are working the same on uh the open AI it's much faster I have seen that it works fine but you see that now it shows me something that autogen repository provide range of tools and various programming languages Great. Can you show me the details? Oops, it's still not showing. Uh, that's a that's a KVR with a local machine that we have got. Maybe uh I can create a new session and ask some other questions. Let's say the same question that I asked with the uh with the manus over here. So, I'm going to go copy the same question. I'm going to paste it

11:37 over here and let's try to ask this question this time to see if this works or not. Um, there we go. It's going to hopefully do that. And look at that. We have got this plan over here. It says, "Search for the popular uh and well-known Mexican restaurant in Oakuckland, New Zealand uh using Bing. Consider factor like review, kusine, quality, and customer rating to select the appropriate establishment and then navigate to the chosen restaurant official website and book uh or a booking platform if available and fill out the reservation forms. So now over here you can also add other steps uh saying also uh book the table by uh 8:00 p.m. uh on Friday. So you can also do that over here. You can keep on adding the steps that use uh that is needed for

12:24 this particular plan to be executed. This is what is this uh the human- centered web agent because you have more control in the agent before the agent itself start performing any of the operation because over here as you can see in the manus this guy creates all the plan for us and uh it will uh execute things for me but over here we as a human has got more control uh and we can also change the UI during this particular operation uh uh and then perform the uh the action over there. That's the power of how we can make use of the magentic UI as opposed to the uh the manus or any other web agents. So over here because we have given all the steps now I can accept the plan and the moment we accept this plan the uh magentic UI is going to start performing

13:10 the operation on the right hand side and then it's going to execute. And one more time just a word of caution because we are running everything within our local machine. these things might be way slower uh than compared to how you do it with um with the actual uh actual open AI APIs or maybe the anthropic API because those are very faster. It runs the inferencing on the on the actual cloud uh than what we are doing over here. Look at that. So now it has went to the Bing. It is going to find for the famous Mexican restaurants uh in Oakuckland. And you'll also notice that pretty much like the computer use the mouse will just hover here and there and it's going to go and click things for me uh and it's going to perform the operation as if like a user and any

13:56 point of time you can take the control and this uh magentic UI is going to wait for the operation to happen. So if I go take the control look at that now the magentic UI is going to say I will wait for you to complete uh before you hand it over to me. So if I go and give it back to the magentic uh UI now the magentic UI takes control uh and then perform the operation. So this is the level of how the entire operation happens uh with the magentic UI asks opposed to um the other web AI agent which is available. So I'm going to say uh continue because now this guy has stopped executing things. So I'll just wait for the entire execution to happen. I can also say that in the step two it says make a reservation of the chosen website. Based

14:42 on the search result we have identified several popular uh restaurant. So this will help you chose the most. Could you please provide any preference? So I'm going to say um maybe traditional because that is what it is asking me to uh give some answer. So if I uh answer that like traditional Mexican restaurant I really wanted to go with. So now it is going to go and process based on the traditional restaurant that it has found. There we go. And because we have given the traditional uh over here the plan has changed. So I'm going to go and accept this plan. Uh and now this plan is accepted. And the magentic UI is going to keep executing based on that plan that it has got right now. And finally you can see that we have got uh the details over here. So based on the search it has found on the trip adviser

15:28 that the highlighted traditional uh restaurant is the Mexican mexical fresh new market. It has got 28 unbiased reviews uh and uh the ratings are not specified and these are the details it has got and it will also saying that next up is it need to go to this uh Mexicland over there you can go and click this this is an official website for booking we can navigate there and then we can also uh book the ticket. Um and if you are using the actual open AAI uh model then we can also say can you go and book it in this particular link it'll also do it for you but I don't think this is going to work within my local machine because it's already cranking the fan sound over there. So maybe I can try but I'm not 100% sure this work or not but we can say can you uh book it for me from the site you have

16:17 mentioned and hopefully it is going to go and do this for me as well. But this is the thing guys. This is how you can see that the magentic UI can do things for you on your behalf. Well, you can just ask some questions and the magentic UI is going to do things uh and it is going to take control of the entire system to do this operation. So because now we have asked to book the new site over there. So see that it has created a new plan and now it is going to do this uh this entire operation for me this time as an entire new operation, new plan altogether. So that is what is going to happen on the uh right hand side over here. Just wait and see what is really going to happen there. So you see that now it's going give me telling me like what I should do. Okay. Uh go for it because it's asking me that uh

17:03 navigate to this particular URL and then do it. So it's going to keep doing it. Yeah, I know this is uh this is not doing quite a lot of different operation. I suppose this thing should have happened by itself and because it is running within my local machine and also very very small model uh it is not handling all these operation for me but um this should ideally be taken care by the uh the magentic UI to do it for me over there. So this is about the magentic UI and I think this is quite awesome where it does things for me uh automatically and you can also see that this is pretty much like the manus which is going to do the same kind of operation. I've also tried out with some other plans over here. For example, I also tried doing the reservation uh over here before and also it did a lot of different search to perform this

17:49 operation and also tried finding the latest publication of the Microsoft research uh over there. Also, it gave me all the links as you can see over here and also tried creating a new login uh for my uh for the website like PBEC and it also did the same exact things over there. So all these operations were actually happening for me uh while I tried the different session. So see how that goes for you like how you can use the magentic UI to perform this operation. But this is quite cool. This is another web agent which is human centered which gives more access to the human uh and it helps you perform the operation and this is yet another tool which I think is quite amazing built on top of autogen. Once again thank you so much for watching this video. Catch you in the next one. [Music]

Share

More video summaries