Video24 min watch

FREE Skyvern Web Browsing AI Agent Beats Claude Computer Use?!🤖 Open Source Browser Automation Local

In this video

  1. 00:00Introduction: Skyvern's New Browser Automation Upgrades
  2. 00:34Overview: Skyvern's API for Manual Task Automation
  3. 00:58Skyvern Features: No-Code, Multi-Platform Integration, and Y Combinator Backing
  4. 01:50Workflow Examples: Automating Invoice Retrieval, Forms, and Job Applications
  5. 02:26Technical Setup: Agents and Browser Automation with Playwright
  6. 03:12Agent Types Explained: Navigation, Data Extraction, and 2FA Agents
  7. 03:42Setting Up Skyvern: Docker Compose Guide
  8. 04:05Configuring Environment Variables and LLM Options
  9. 04:41Finalizing Setup: OpenAI and Anthropic API Key Integration
  10. 05:19Starting Skyvern: Docker Containers and Dashboard Overview
  11. 06:05Running Tasks: YouTube Subscriber Count Example
  12. 06:32Task Management and Diagnostic Tools
  13. 08:02Accessing Task Data: Parameters, HTML, and API Integration
  14. 09:24Introducing Stride Agents: AI for 24/7 Appointment Setting
  15. 09:59Using Skyvern Templates: GEICO Insurance Quote Example
  16. 10:52Live Task Execution: Tracking Progress and Logs in Docker
  17. 11:26Task Accuracy: Auto-Fill and Dummy Data Management
  18. 12:41Workflow Automation: Multi-Step Processes and Task Management
  19. 13:38Building Workflows: Task, Loop, and Parameter Configuration
  20. 15:05Advanced Workflow Blocks: Text Prompt, Email, and File Parsing
  21. 15:27Running Workflows: Real-Time Task Execution and Tracking
  22. 15:52Error Handling: Adjusting Prompts for Task Precision
  23. 16:26Final Thoughts on Skyvern's Automation Potential
  24. 16:52Comparison with Claude Sonnet and Other AI Tools
  25. 17:22Community and Resources: Join the Stride Community
  26. 17:44Conclusion: Daily AI, Automation, and Business Content on the Channel

Links from the video

Transcript

3,596 words

What's going on everybody Josh pook here and recently anthropic just released claw Sonet 3.5 computer use and we've done videos on this channel covering how to actually set it up using their uh specific demo version as well as I showed you how to use agent.exe so if you want to check out those specific videos you can check those out but in today's video we're going to be looking at a tool that has been around for a while and recently has gotten

Some really really cool upgrades skyver automate browser based workflows using llms and computer vision skyver provides a simple API endpoint to fully automate manual workflows on large numbers of websites replacing brittle and unreliable automation Solutions we're going to show you how you can set it up and self host it for free let's dive right into [Music] it all right guys so here is the official skyver website I'll leave links to everything we cover in the description down below here is a

Example of skyver in action basically you create specific tasks using prompts and then skyver is going to start taking those actions right here as you can see and you'll be able to see a specific preview of that job being done such as filling in forms doing uh tedious manual tasks it is backed by y combinator works on any website powered by natural language apid driven an infinite scaling run thousands simultaneously and no Code and low code options features you won't find

Anywhere else such as capture support two- Factor authentification proxy networks explainable AI data extraction so extract data from workflows in any scheme of your choice including CSV or Json so a few different test examples would be streamline invoice retrieval submit Dynamic multi-step forms automate procurement pipelines apply to draw with one click Brees through government forms complete workflows in any language also too they have some pretty cool Integrations so they have P connect zapier um a few of these are actually in

Review such as work Kau make.com impr progress n8n and then fresh sales I'm just going to quickly go over how skyver works so it was inspired by the task driven autonomous agent design popularized by baby AGI and auto AGI with one major bonus we give Skyrim the ability to interact with websites using browser automation libraries like playwright so skyern uses a swarm of agents to comprehend a website and plan and execute its actions so it has the interactable elements agent so

This agent is responsible for parsing the HTML of a website and extracting the interactable elements we have the navigation agent so this agent is responsible for planning the navigation to complete a task examples include clicking a button inserting text selecting options Etc then we have the data extraction agent so this agent is responsible for extracting data dat from a website it's capable of reading the tables and texts on a page and extracting the output in a userdefined structured format and we

Have the password agent this agent is responsible for filling out passwords on forms or websites we have the two-factor authentification agent so this agent is responsible for filling out two- Factor authentification on a website and then we have the dynamic autocomplete agent so this agent is responsible for filling out Dynamic autocomplete forms on a website so a few examples would be address forms University dropdowns and more and here is the example of how it would actually perform a task such as

Go to Geico and generate an insurance quote so we have breaking it down into six different steps and then executing on those specific steps with different agents and you can watch the demo here as well made by one of the founders I believe now there's a couple different ways to set it up there's the docker compos setup which is recommended or the full setup for contributors we're just going to use the docker compos setup because it's easiest so all you're simply

Going to do is you're going to start off by Clon this repo right here so you're going to go into your command terminal and you're simply going to run get clone okay once you run that you're going to Simply change directory into skyver like so and then what you're going to want to do is you're going to want to run code dot to open up the repository in something like VSS code or you can open it up in cursor whatever you

Want that's going to open up the repo like this and you're going to want to go to the docker composed. yaml file the main thing you would want to do is scroll down right to here here where you'll see the different environment variables so we have the open AI API key right here so if you're going to use open AI API key then you want to paste in the API key here if you're going to use another llm provider then you

Can enable it like so by doing enable anthropic true and then uncommenting this one too so llm key anthropic CLA Opus now if you scroll all the way down to here you'll actually see the supported llms as well as the environment variables you can see here enable AI true false enable anthropic true false you can see llm key and the supported versions right here for llm so um you could do these ones right here and throp the cloud Sonet Sonet 3.5

Whatever um GPT 40 Etc so if you are going to change some stuff around you can scroll down there to customize it okay once you're done that you're going to need doer. comom as well so that's one prerequisite um so just go to doer. comom if you don't have that already and then you're going to make sure you're you have that opened on your desktop and then simply run Docker compose up that is going to start up these containers right here

Once these get spun up you'll be able to access skym on Local Host Port 880 all right so this is what the skyver dashboard looks like you can see here we can create new tasks we can check out our previous tasks we can check out our workflows right here and we can go into our settings so in our settings we can see our environments we're in local we can see our organization and we can see our API key if needed all

Right you simply click reveal then you'll have it there and then here are the workflows right here we'll dive into more of these in just a second okay so all right so we're going to start off with the prompts go to YouTube search for Josh pook and get the subscriber count all right we're going to run this all right so the task is done running you're going to go to your task right here so when the task is running you'll see

It right here you also see your Q task then you'll see your task history so we're going to click on task history right here we can see the ID the URL when it was created and the completion status um once we click here we'll see the extracted information so we can see that it did get the subscriber count 9,390 subscribers and we can see the specific actions that it took so we can see here the search bar is empty inputting Josh

Poock will initiate the search process so it went to YouTube here it took a screenshot and then the llm was able to read that we can see here step two clicking the search button will execute the search for Josh pook we got you know the channel right here and then it clicked on the channel link to navigate to Josh poot Channel and then the user goal is achieved as a channel page is visible extracting the subscriber count is the final step

And then boom so even if you go here you'll see a recording so sometimes this works right here it's I guess it didn't work it's like kind of white right here but sometimes you do get a recording um or you should and then parameters so you can see here this is the task URL and input parameters so got the URL here created at navigation goal we got nav ation payload data extraction goal extracted information schema you can go to Diagnostics here

You can see the different steps all right step one step two step three and we have info right here we have annotated screenshots we have action screenshots right here all right element Tre we can see the different elements right here that it got we can see the HTML element Tre that it got right here we can see the prompt right here the full entire prompt make sure you output valid Json we can see the action list right here we can see

The HTML raw so all the HTML that it got we can see the llm request raw okay so you can literally see every single thing pretty much related to this task so you can copy this curl command to your clipboard and then if you paste this into your um command line you can actually run this task in your command line right so it will copy and paste all the different uh variables and everything you need in terms of your API key

The specific task ID the website URL the API endpoint all that different stuff and you can actually run these in um you can use the API essentially right so um pretty cool stuff if you do want to use more of the a API I would suggest checking out the docs here so the docs are pretty in-depth um if you go down here to tasks API you can see the uh how to initiate a task the the different headers the body right

Here the different variables retitle 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 El performed 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 all right you'll also see here that there are skyver templates so you can go ahead and try some of these templates out and

Then you could also create your own templates here so you can see here we got Josh pook YouTube subscribers you could create your own templates if we go to skyvern templates and run this one generate an insurance quote you from GEICO we could run this we can see the URL the navigation goal we can see advanced settings right here which will show the payload all the different payload stuff and parameters we could go to extraction we can see the extraction goal

The data schema we could go to advanced settings we can see the max steps override the web hook call back URL error messages two Factor uh verification URL two Factor authentification ID identifier you can even see the curl right here as well as the Run button so if we go ahead and run this you can see here task has been created so if we go to our task here we can now see that this task is currently being run so we

Can actually see it loading up Geico right here live which is pretty damn cool we can see it's running and it will progress as things go forward you can see here that it's loading the page so if we check in our Docker containers right here we can actually see what is going on we can actually see the feed and the logs of you know scraping website updating steps uh starting agent successfully went to Geico all the different cool steps here so

You can track that as it's running okay so we can see here that it just entered in a zip code and now it is literally filling out the form it clicked on auto okay so we can see the steps here getting filled out so the ZIP code is required it just entered it in Select Auto to ensure the quote is for auto insurance not home insurance click Start my quote to proceed with generating the auto insurance quote right now it looks

Like our page is loading right here right it's asking us for our first name last name and date of birth right so it's just filling out some dummy data right here so it has Chris for the first name last name it's putting bacon put some fake date of birth right here now it's asking for our address so I have to say this thing is pretty good like it's getting everything correct not misclicking anything so far and it's intelligent enough to put

In dummy data all right so we can see here it's filling out Garfield Avenue loading the next page yeah and the good thing about this is that it's infinitely scalable you can run these tasks and you can run multiple different tasks using a computer and it will perform these tasks when we were using Claud computer use we were having issues actually filling out a form on my website to book a call and so far with skyver this is like a 18-step

Um task right here we're at so far in terms of a form it's actually going to be more and it's literally had no issues it's gone through everything it's getting it correct all right so it was 64 total actions and it actually did up failing just because of a few different things for the application of it entering dummy data that like it wouldn't just let it enter dummy data in the form but other than that it I mean it was pretty

Long task as you can see the screen recording is right here um you know it asks for things like social security number and stuff so obviously um it's not going to actually be able to just get a quote from using dummy data here but we can see all the information so the next thing is the workflow so if we go to workflows here so you can create workflows you'll be able to see your workflow runs if we click over here into

A workflow we can see the pass runs we can edit a workflow so this is like an editing um space right here kind of similar to something like you know n8n or make.com it's using react flow so if we click here we can add different nodes we have task block for loop block text prompt block send email block file parser block download block upload block all these different actions we can take so for example if we do a task block right

Here I can name this as get subscribers I could put in the URL you could do basic or advanc so if you do Advance right here you'll get some extra things such as extraction limits you know Max retries two Factor authentification right here right so pretty cool stuff we'll just stick to basic for now you can add parameters right here so let's say like we wanted subscriber count to add a parameter you can click up here and then click on ADD

Param so you can click on workflow parameters uh credential parameters uh context parameters all these different parameter options say if you clicked on right here you can add the key description Source parameter if needed or you can go here workflow parameter you can change it to string number Boolean file Json Etc you can click up here credential parameter so key description URL parameter key collection ID all these different options and then a workflow is essentially allowing you to run more multiple

Tasks at Once In Like A workflow right so if I wanted to get subscribers I could also add a task block right here to do a next action such as you know get the view count of the most recent video or after it goes to to the YouTube and gets you know the subscriber count then get the you know time zone I could say go to Google and search for time zone and weather in Sydney Australia all right you also could

Use for Loop blocks which will essentially repeat these uh tasks in the block right here so you can you could add um a task like this within the block you could add additional nodes inside this block as well and it will repeat and then we have additional things like I said text prompt you could go here add some prompt add a data schema if needed you go here and I'm not going to go through all these other different nodes right here

But you can make some pretty decently Advanced uh workflows here right so you can save it you can rename it if needed so we'll save this test prom uh workflow right here which has two steps and we can go ahead and run it actually we wouldn't need that specific parameter I had so I'm just going to rerun it now right so I'm going to run the workflow and the workflow has started if I go back to workflows now I can see

Workflow runs and we can see that this is running right now if I click on it it opens up the workflow and we can actually see the workflow running live so we can see it going to YouTube right now it's going to make the search also too if you copy this Cur right here it will copy the command for you to run the workflow all right now we can see it is searching for the time zone and weather in Sydney Australia

On Google so if we scroll down we can see that the task has been complete and now it is running the Google task all right so we can click into the individual tasks and see the different tasks being complete okay for some reason the Google task didn't work if I click inside I can see here that the search query is already entered clicking the Google search button should run the search previous attempt failed but should try again so it timed out

I believe I think it was clicking this button which wasn't searching it so you may have to be a little bit more specific with your prompts as well um you know pertaining to what website you're using and how it's specifically structured uh you know for the screenshots and everything but uh that's pretty much it for this video guys I think you get the gist between the workflows the tasks and all in all I think it's a really cool tool I think

It's even cooler if you actually start leveraging the API side of things and you can actually scale automation different tasks you could build predefined templates for your specific use case and I think you could actually make um some really cool things happen with this let me know what your thoughts on in the comments down below let me know how you think this compares to Claud sonnet's computer use and some of the other tools we also covered on this channel such as

Agent Q agent e if you haven't seen those videos check those out on my channel as well to see what is the best browser based automation tool leveraging AI other than that guys if you're new here we upload videos every single day on AI automation Business Marketing sales Etc so if you like that type of content and you got some value here make sure to smash the like button comment down below 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 so you can join our free Facebook group and Discord Channel and then if you want help growing your business and accelerating your stride book a call with the link down below to speak to myself we can go over our custom growth plan for your specific business and see if there's any way to scale your business by implementing AI systems and Cutting Edge

Marketing strategies 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