Introducing AnythingLLM
Hey everyone, my name is Timothy Carbat. I'm the founder and creator of AnythingLLM. If you're not familiar with AnythingLLM, we are an all-in-one AI application that just runs entirely on your desktop or your laptop where you can run local models to just chat with. You can also upload docs. You can chat with images. You can also create and run AI agents to either accomplish tasks or do automations for you.
All of this local for free. And we also have an open-source part of our whole project that is MIT licensed. So if you want to follow us there, we would greatly appreciate that.
Why Context and Web Scraping Matter for AI
If you clicked on this video, you're interested in web scraping. Now, web scraping is not a new concept. It's been around for probably as long as the internet has. And in the current AI information age that we've all kind of fell backwards into, one thing has become really, really clear. Context is key. If you want your LLM to give you a factual answer or just one that has some kind of grain of truth or real-time information because models don't change their weights, you have to rely on usually online sources.
Now, context can also be documents and PDFs, but for the sake of this video, we're talking about online web links.
The Big Problem: Scraping Behind a Login
Now, there is a whole ecosystem around web scraping. If you're a developer, you're probably already familiar with a lot of these tools. Selenium is one of the oldest ones just for doing automations or web scraping. There's Puppeteer as well. And if you're maybe more attuned to AI solutions, there are open-source projects like Firecrawl and Crawl for AI, both of which are libraries to help developers crawl the web.
If you've used ChatGPT pretty much since it came out or Gemini or Anthropic, you know about web scraping because when you when you give your LLM a link and it goes on the internet and gets that information, that is web scraping. But every single tool has one big problem. Only half of the internet is actually unlocked. Some of the most important and contextually aware online information is typically hidden behind a login screen. As a simple example today, I'm going to show you some private information that I can normally just access in my browser and we're actually going to give that context to my LLM to help it do something or form an answer or just help me out.
The Solution: Authenticated Scraping in AnythingLLM
What I'm talking about today is that as of anything LLM desktop 1.8.3, which just came out today, you can now do authenticated web scraping. Again, everything about this still on your device, still private, and it works for Mac, Windows, and Linux. Any website at all that you can log into in a normal browser, you can do so in anything LLM, and your LLM can now use that information in real time and basically like log in as you.
And today, I want to show you a very simple example of using private authenticated information for web scraping. Keep in mind this is on top of scraping the entire internet. So now, not only do you get the general information available on the internet, you also get that really niche private stuff that is honestly probably more useful to you in some situations depending on what you need.
Demo: Trying to Scrape Private LinkedIn Data
So all I've done is I've just downloaded anything LLM for desktop 1.8.3 here on my old Intel MacBook because that's just what I have right now. We're just going to run an example of when you need authenticated information, but you can't get it. So, an example of authenticated information that could be useful. Again, just think of this as any page behind a login or a paywall that you already can access in your browser. Here are my LinkedIn notifications. Now, for the most part, everybody knows LinkedIn notifications are pretty useless, but asking an AI to go and check and say, "Hey, is there anything useful or did anyone cool follow me?" That's not something that an AI could actually do for you today anywhere. Try and ask ChatGBT or Anthropic to do this and they can't do it.
So in anything LLM, we have a whole bunch of agent skills and I just want to show that scraping websites is on. It is one of the features that we have. It's built in and it runs again privately on your machine. But if we were to ask our agent to do something like, hey, can you scrape LinkedIn notifications and then tell me if anyone cool is following me? Let's see what happens.
So, here I've typed out a prompt and it just says, "Hey agent, check my LinkedIn notifications and tell me about any updates you think are relevant or useful to highlight for me." And because most models like to throw a refusal if it's not a tool call, I even gave it the actual link that I would go to in my browser just to see my notifications. And you can see it gives us a response that says, "I'm unable to access it because it requires logging in." This is legitimately what I just said is one of the main problems with using a lot of the tools that exist out there today.
So now let me show you anything LLM's authenticated web scraping tool. It's built right in. It's free. It's not a plugin or anything like that. All you do is just click on settings and then you want to click on browser tool and then we have this open private browser. And what this does is it opens up a little sub window. It looks like a browser. It's actually part of the anything LLM app. But one of the quick example links that we have in here is LinkedIn. If we go to this site or if we were to go to that same URL, you can see we are also hit with a login page. But all we need to do here is just log in.
And so I've already gone past the login page. I filled in my information so that people don't spam my email. And now it's asking me to sign in. So I'm going to just open up the LinkedIn app on my phone. Click yes, it's me. And now we're on LinkedIn. Now, there's really nothing more for me to do here. I can actually close the browser, but you can see that now the URL we asked it to scrape is now available to us. So, what we can do is let's just close this because we're done.
Demo: Successful Authenticated Scraping
And now let's see what happens when I just rerun that same request. We don't have streaming enabled on tools, which is why it just kind of popped up. But you can see that it said it was scraping the content using the web scraper, except now it says that Martin K and two others started following me, which is great for my network. Bunch of people reacted to the anything LLM mobile app post that I made, which by the way, if you're interested in a completely ondevice AI agent experience, much like what I'm showing you right here, then we're actually launching a mobile app and you can actually well, I'll put the link in the description so you can sign up for the beta.
There are also some industry updates which LinkedIn loves to spam. Great to know that they're there and obviously you can see that you know our company page appeared in 453 searches which is great. So this is information and of course I know this is a very small sample but think about all of the content like your Gmail or a document that's password protected that you access on the web and a company internet page that's behind SSO. All of these things are impossible to collect or interact with for pretty much any other AI solution you're using.
Integrating Authenticated Scraping into Workflows
This is why this feature is so exciting. One thing that's worth highlighting that I did in this example was I just gave it that link. But giving it the link is only part of it. If I were to put this link in our web scraper right here just to use it as a document, a static document, you can do that. If you want to use this authenticated web scraping in agent flows, actually you can do that in the web scraping block. All of these things are fully possible for you to do. And it's all just one platform, one unified experience that you can use to have an LLM just be smarter and use more personalized context when answering your questions.
Conclusion & What's Next for AnythingLLM
Hopefully, this video has given you a small glimpse into how powerful AI models can become when not only they have access to the internet, but they have the access to your internet. Obviously, with your permissions and with your security in mind, all running on your device. Anything LLM's authenticated web scraping tool is actually just the beginning of a few key release features that are going to be coming out on anything LLM desktop soon. This is just a really exciting starting point.
If you really are interested in these kind of ondevice experiences that just work for you and are more productivity focused as opposed to developer focused, then download anything LLM today. Follow us, join our discord, star us on GitHub, whatever you want to do. We will stay in touch.
