Chat with any webpage using your OpenAI key

The LLM context transfer problem

Ever since the launch of ChatGpt I have been avid user. I’ll use it for everyday simple questions (like how long does an Avocado stay fresh after being cut open?), help with writing documents at work, programming assistance, summarizing long articles, and more. The list is endless.

Even though ChatGpt is no human, there’s no doubt about its usefulness for simple tasks. I like to think of it has a smart middle school kid that can follow instructions nicely and work with the data it is given.

One annoying thing about using ChatGpt a lot is that you spend a lot of time giving it the context it needs to answer your question. This can involve a lot of copy pasting into it. We’re seeing tools being released every day that try to solve this context transfer problem for you. For example, if you’re a software engineer then you’re probably familiar with https://www.cursor.com/ that basically integrates LLMs into your IDE to make the context transfer problem easier.

Solving context transfer for browsers

I wish there was a tool that integrated LLMs into web browsers the way Cursor does for IDEs. After some searching, I found the Arc browser’s chat with a webpage feature and started using it. There was one problem though — I still use chrome as my primary browser and not Arc. So I went looking for chrome extensions that would let you do this. I found https://sider.ai/ and similar tools that will give you a ChatGpt like window embedded in your browser which makes context transfer easier but doesn’t completely solve it. I was still copy pasting a bunch of stuff into that small window from the webpage I was on, so I continued searching.

Finally, I stumbled upon https://rockyai.me/ which was exactly what I was looking for. It seamlessly takes the context of the current open tab you have in chrome and lets you chat with it. It is similar to Arc browser’s chat with webpage feature but for Chrome with a better UI. It also respects your privacy in that it’ll only access the contents of your webpage once you interact with the extension.

The best feature — you can use your own OpenAI key with it. For someone like me who is paranoid about sending the webpage content I’m browsing to someone else’s servers, this was super helpful because I trust OpenAI more with that data as a big company than something smaller.

RockyAI and Privacy

I inspected the network requests the extension made to ensure it wasn’t sending any personal webpage content over the wire while using my own OpenAI key. Here’s what I found -

1. The extension makes two requests every time I ask a question. The first one is a request to the OpenAI v1/chat/completions endpoint with the questions and relevant context of the webpage.

2. The second request is metrics tracking request to Rocky AI. It sends the tab hostname request to their servers.

While I wish the tab host name metric collection was optional, this information isn’t sensitive and I don’t mind sharing it.

Conclusion

I have been using RockyAI for some time now to summarize articles, get summaries of HN comments on threads, and find relevant information in long developer docs, among other tasks. I love the ability to use my OpenAI key with it, which adds an extra layer of privacy.

I’m genuinely excited about the potential of integrating LLMs into web browsers, and I believe tools like RockyAI are just the beginning. If you have recommendations for other tools or extensions that could further enhance this browsing experience, I would love to hear your thoughts in the comments below!