Video14 min watch

NEW Agent.exe Claude Computer Use AI Agent UI! 🤖 Open Source - Controls Computer Claude Sonnet 3.5

In this video

  1. 00:00Introduction: Anthropic's Computer Use Feature Overview
  2. 00:47Introduction to Agent.exe: Claude's New Computer Control Capabilities
  3. 01:25Tweet from Kyle Corbett: Launch of Agent.exe
  4. 01:59Agent.exe Interface Overview: Example Task Demo
  5. 02:28Comparing Original vs Forked Version of Agent.exe
  6. 03:04Setting Up Agent.exe: GitHub Repo, Installation, and Configuration
  7. 03:54Forked Version Features: Send Button and Chat Continuity
  8. 04:24Stride Agents AI: Commercial Use Case Promotion
  9. 05:38Example Prompt Demo: Checking YouTube Subscriber Data
  10. 06:26Prompt Demo: Fetching Weather and Time for Sweden
  11. 07:22Task Demo: Creating a P&L Spreadsheet in Google Sheets
  12. 08:31Addressing Issues with Browser Control and Tab Functions
  13. 09:05Final Thoughts: Limitations and Potential of Claude's Computer Use
  14. 09:45Closing Remarks: Testing Agent.exe and Community Links

Links from the video

Transcript

2,172 words

What's going on guys Josh pook here and I'm sure most of you have already heard of anthropics new computer use feature in fact we actually did a video on it yesterday showing you how you can set it up very very simple in a Docker container and I'm sure lots of you have seen this in action either on my Channel or on Twitter or wherever where people are using this claw computer used demo that anthropic provided that is using a virtual environment

Right here using no VNC and a chat interface on the left to control browsers or control the computer and do different actions and this is good for testing and it definitely has some pros such as being more safe but it definitely has a few downsides such as being really slow and really not a practical application of this new feature so of course we could go ahead and start implementing the code here and build something ourselves but what if there was a

Way where we could easily set up a nice pre-built user interface that would actually do these actions for us on our own computer well that's where agent.exe comes in the easiest way to let claud's new computer use capabilities take over your computer it's free it's open source and I'm going to show you how you can set it up in today's video as well as I'm going to give you access to my forked version of it with a couple new enhanced features

Let's dive right into [Music] it all right guys so I was on Twitter today and I saw this gentleman here Kyle Corbett tweet out just launched agent.exe a free open-source Mac Linux Windows app that lets you use Claude 3.5 Sonet to control your computer this was a fun little project to explore the API and see what the model can do computer use is really cool I expect 2025 will be the year of agents and yes I can definitely agree with that

Prediction so here's what the user interface looks like here it's a little chat window and you can see here it has an input box right here where we can give it our prompt it has a delete button it has a let's go button and we have kind of the history interface right here here's an example of it in action so you can see here it's running right now and you can see that it's finding flights for Seattle for next Tuesday or

To Wednesday and you can see there's a pause button here and yeah it's putting San Francisco here we're not going to watch the whole thing cuz I'm about to demo it for you in just a second but you get the gist okay so you can go ahead you have really two options so you have two options you can use the original version right here which is really good but I personally found there was a couple little features that I wanted it

To have such as the ability to when I pause the chat to be able to reprompt the chat with a new prompt or a new message in the chat without it starting a complete new chat as well as I personally found for me running it on a Windows using my specific browser I had a few issues with the clicking you know it not being able to click in the exact place it was supposed to so you could try his version if

You are running into any of those issues then you can try my forked version of it okay so both the repos will be linked Down Below in the description and all you're simply going to do is you're actually just going to copy the repo right here the URL you're going to come into your terminal and you're going to run get clone and then the repo URL once you do that you're going to change directory into agent.exe okay from that point on

You can run npm install Okay so that's going to install it right here and then you're going to want to run code dot or cursor dot you're basically going to want to open this in vs code or cursor okay once you have it opened in your editor of choice you're going to see that there's a EnV example file you're just going to want to clone that and change it to EnV and then you're going to want to paste in your anthropic

API key all right if you scroll down in the repo 2 you'll see those exact instructions here of course if you're using my Fork version you would just change out this URL right here and then all you're going to do is run npm start so we're going to run this right here okay and here we have our agent.exe now you will notice that the fork version right here has a really it's just has one different button right here so it has

A send button and a new button so a new button is going to create a new chat and the send button is going to keep the messages within the same chat all right and this is still obviously a very new tool as well as computer use is obviously a very new tool so there may be some issues maybe some bugs as well as the fork version I made it's definitely not perfect by any means I just wanted to add some additional

Little features here just for my personal use and maybe some of you guys will enjoy that too are you tired of pouring thousands of dollars into appointment Setters only to watch leads slip away imagine having a team of elite sales agents booking qualified appointments for you around the clock no more wasted time on training no more frustration with performance and no more draining your budget on inconsistent and expensive call centers introducing stride agents AI powered appointment centers that work 24/7 never

Get tired and book appointments while you sleep trained on thousands of successful conversations our AI agents outperform human teams at just oneth of the cost join the ranks of businesses that doubled their appointments and booking rates in just a matter of weeks don't get left behind in the AI Revolution visit stride agents.com now and transform your entire sales process with Cutting Edge AI technology it's time to accelerate your stride with AI agents now you can use this on any browser but

Kyle says that Claude really likes Firefox so we'll just go ahead and run it on Firefox even though Zen browser the one I was using is based off of Firefox all right so I'm going to say search Josh pook on YouTube tell me how many subscribers he has and the names of his last three videos and their views so let's go ahead and send this all right so it says I can see Firefox is open let me click on the YouTube

Icon and search for Josh pook so it's loading up YouTube okay so it pulled up my channel here okay so it made a mistake and I just told it that wasn't the LA the name of the last two and as you can see it's continued on with the same conversation and remained context within this whole thread as we can see here it says I can see Josh po has 9.39k subscribers according to the channel page and then we can see the

Last three videos is one new Claude AI agent computer use which is correct and then it gave me two different wrong names right here so I let it know that that wasn't the name of the last two and it said my apologies let me scroll back up and get the correct video and then it said Claude AI agent computer Ed and it said fre fi data AI agent which is true one day ago and then building a custom AI agent memory

Tool with python okay so this is incorrect so I got two of them the last one is incorrect I could tell it to get the last one correct but uh we'll just try a different prompt here all right so now we're going to try get me the current time and weather in Sweden just any place there and I'm going to click on new so we can create a new thread okay so it's picking stockolm Sweden and it's going to get us

The time here all right searching for weather Stockholm Sweden okay and boom so current time in Sweden Stockholm is 8: a.m. current weather sunny temperature 10° C wind etc etc etc so did that correctly very easily all right so I'm going to say open a new tab go to Google Sheets and enter some dummy data with formulas around this month's p&l okay so it's just going to Sheets google.com so it had some trouble opening up a new tab there okay so

It just it's messing up right now I'm actually going to pause it I'm just going to say you didn't create a new spreadsheet send it okay I just had to restock up on some credits because one thing about computer use is it blows through credits quick so keep that in mind we're going to just send this okay so at least now it's creating a new blank spreadsheet okay we got the spreadsheet here let's see if it can actually input some data

When you're doing it make sure that you click on your uh browser so that when it presses a key like controlt for a new tab or enter that it actually works um so I'm just it wasn't working for me so I'm just going to say sorry restart the pnl and send it okay so I'm saying restart the p&l make sure you click to the next Tab and don't type things in the same tab take screenshots after each so you can make

Sure you're doing it correctly going to hit send okay so it is doing some data entry here so I was having some issues before with the tab uh button not working so I actually had to go back in and fix it in the code but we can see here that it is working now the one thing is the revenue the capital um and lowercase is kind of messed up just because I had caps on and then it piped like that so

Okay so it is still kind of messing up a little bit it's not perfect there okay so it's not perfect as you can see I'm just going to stop it right there you know but it was going to keep going um and like I said before guys I said this in the last video I'll say it in this video obviously claw computer use isn't perfect obviously this agent.exe isn't perfect but I think this is a really cool tool for at least

Being able to easily try it out on your own computer you can go ahead use this for free it's open source and at least you won't have to deal with the slowness of using it in that virtual environment the one that Claude provides as a demo so other than that guys I don't want to make this video too too long I did more tests in yesterday's video this obviously is the same tool so it's going to have similar success with certain

Tasks but now we have it in a nice little UI interface which is directly on our personal computer so you can go ahead use this try it out for different tasks see how it works and let me know in the comments down below what you think like I said I'll leave the link to the original repo down below I'll also leave the link to my fork and like I said my Fork isn't perfect I just made a few slight adjustments just

To get a few things that I personally wanted so you can use whatever one you want or maybe test out both other than that guys if you're new to the channel we upload videos every single day on AI automation Marketing sales business growth if you're new to the channel and you like that type of content make sure to leave a comment down below smash a like button and subscribe to stay up to date with the daily uploads also too guys if

You haven't already joined our free community stri community.com I'll leave a link down below to our free Facebook group and Discord Channel other than that guys I will see you in tomorrow's video keep hustling keep grinding and of course guys accelerate your stride take care

Latest from the blog

View all