VisionGPT

VisionGPT
Photo by Paul Skorupskas / Unsplash

VisionGPT is a Chrome extension that lets you capture part of a webpage and ask AI to analyse it.

The idea is pretty simple.

Instead of taking a screenshot, uploading it to ChatGPT, and explaining what you want to know, why not do it all directly in the browser?

That's basically what VisionGPT does.

How It Works

You select an area of a webpage, capture it, and let GPT Vision analyse what's on the screen.

It could be a dashboard, a UI design, an error message, or just something you want AI to explain.

I also added a few features along the way:

  • Custom prompts for different tasks
  • A sidebar to view previous results
  • Local storage for your screenshot history
  • Export options
  • Bring Your Own Key (BYOK), so you use your own OpenAI API key

I wanted to keep the BYOK approach.

Instead of running another backend to manage AI usage and charging users for tokens, VisionGPT connects directly to OpenAI.

Less infrastructure for me to maintain, and users have control over their own API usage.

Building and Improving It

Like many of my side projects, building the first version wasn't really the hardest part.

I got VisionGPT working and published it as a Chrome extension.

Later, I spent more time polishing the experience.

The screenshot capture became faster, the sidebar became resizable, and I improved the result viewer with fullscreen support and adjustable font sizes.

I also fixed some annoying bugs, including screenshots turning black on certain websites.

Not exactly exciting features to talk about, but those small things matter when you're trying to make a product usable.

Eventually, I continued improving VisionGPT, including releasing version 2.0.

I thought it was in a pretty decent state.

Trying Different Licensing Models

One thing I struggled with was figuring out how to monetise VisionGPT.

Initially, I introduced a paid license model.

I thought it made sense. Users bring their own OpenAI API key, and they pay a one-time fee to unlock the extension.

But it didn't really work.

I think I sold only a few copies.

Later, I switched to a free license model, hoping removing the payment requirement would encourage more people to use it.

That didn't really work either.

So when I released VisionGPT 2.0, I removed the licensing system completely.

No more license activation or registration. Just install the extension, provide your own API key, and use it.

Looking back, I probably spent more time experimenting with the licensing model than I should have.

But I guess that's how you learn.

Sometimes the problem isn't the pricing or licensing model. Maybe the product just isn't solving a big enough problem for enough people.

The Harder Part — Getting People to Use It

I also tried promoting VisionGPT, including during Black Friday.

But the response was disappointing.

Even after removing the license requirement, it didn't really gain much traction.

And AI tools were changing pretty quickly.

What felt like a useful standalone extension could become just another feature inside a larger AI product.

I guess that's one of the challenges of building small AI tools.

You can build something useful, but finding enough people who actually need it is a completely different problem.

A Product I Don't Even Use Myself

There's actually another reason why I'm slowly giving up on VisionGPT.

I don't use it myself.

Not because it's broken or doesn't work, but I just don't find myself needing it in my daily workflow.

And I think that's a pretty big problem.

When you're building a product that you personally use every day, it's easier to stay motivated. You notice bugs, discover missing features, and naturally think about how to improve it.

But when you don't even use your own product, it's tough to find a reason to keep maintaining it.

Especially when you don't have many users or sales to push you forward.

I guess I built VisionGPT because I thought it was a useful idea, not because I personally needed it.

There's nothing wrong with that, but as a solo developer juggling multiple products, it's hard to justify spending more time on something I don't use.

That's probably the biggest lesson I learned from VisionGPT.

So, What's Next?

VisionGPT is still around, but I'm no longer actively developing it.

With the changes to the Chrome Web Store, I'm now limited to having only four extensions. Most of my extensions don't get much traffic anyway, so I might eventually have to decide which ones are worth keeping.

And VisionGPT could be the next one to go.

For now, it's still hanging there.

I have other products that need more attention, and as a solo builder, I can't keep maintaining everything at once.

Looking back, I don't regret building it.

I learned more about Chrome extensions, working with vision models, handling screenshots, and packaging a small AI tool into a product.

I also learned that changing a product from paid to free doesn't automatically bring in users.

And perhaps more importantly, building something that sounds useful is very different from building something you actually need.

Not everything you ship will become a business.

Sometimes you build something, put it out there, try a few different approaches, learn from it, and move on to the next thing.

So yeah, VisionGPT is still available for now. Not actively maintained, and maybe not for much longer.

We'll see how it goes.

VisionGPT 2.0 | AI Screenshot Analysis in Chrome
Capture any part of a page and send it straight to OpenAI with your own key. Each saved prompt keeps its model and image detail.