跳到正文
DEV Community· Richard Shaju·· 6 小時前AI 評分42

Toolchat 的故事:獲 450 名用戶、收入為零後關閉

I Built an AI Tool, Got 450 Users, Made $0, and Shut It Down. Here's the Story of Toolchat.

AI 導讀

作者打造的 AI 檔案處理工具 Toolchat 累積 450 名用戶、700 多次請求,卻沒有付費訂戶,最終關閉。工具支援批次處理圖片、音訊、影片及 PDF 等檔案;隨着 ChatGPT 和 Claude 的能力提升,加上用戶難以為額外訂閲付費,作者認為產品難以持續。

正文

I Built an AI Tool, Got 450 Users, Made $0, and Shut It Down. Here's the Story of Toolchat.

Every builder has that one project. The one you were sure about. The one you explained to everyone who'd listen.

For me, that was Toolchat.

How it started

In 2025 I had a simple idea: what if you could process any file just by typing what you want?

"Merge these PDFs."
"Trim this audio."
"Split this image."

No searching for the right tool or clicking through ten menus. You'd just say it, and it would be done.

At the time, ChatGPT and Claude couldn't really do this well. If you asked them to merge a PDF, they'd write you some Python code and try to run it. Sometimes that worked and sometimes it didn't. It never felt like a tool. It felt like a workaround.

So I thought, why not build the actual tool?

I started working on it in May 2025. Just me.

Building it the hard way first

My first plan was to build my own ML model to understand what the user wanted. I built it, and honestly, it worked well. I was proud of that.

But at that stage Toolchat could only handle a few image functions. Every time I wanted to add a new feature, I had to retrain the model. Every single time. It was slow and painful, and I knew it wouldn't scale.

So I made the switch to an LLM. Gemini was offering free API tokens back then, so I started integrating that. Suddenly adding features got much easier.

In September 2025, Toolchat went live.

After that I kept adding more. By the end it could compress, convert, resize, split and merge images, and remove backgrounds with AI. It could trim, merge and compress audio and video, and pull the audio out of a video. It could merge, split, lock and unlock PDFs, convert Word and Excel files to PDF, and even turn a PDF into dark mode for night reading. All you had to do was type what you wanted, and you could do it for a whole batch of files at once.

Then I added a subscription plan. It started to feel like a real product.

The part where people actually showed up

I did some small-scale marketing: AI-generated ad videos and some influencer content. And it worked. People came in, and most of them actually tried it at least once.

The Twitter split image feature became the surprise hit. It got more use than anything else and even pulled in some organic users on its own. I hadn't planned that at all.

Here's where Toolchat ended up:

  • 450 users
  • 700+ requests
  • Users from 10+ countries
  • 0 paid subscribers

Yes, zero.

Meanwhile, the world moved

While I was building, the AI world was moving very fast. The same tools I was trying to improve on, ChatGPT and Claude, got smart. Really smart. The things they struggled with when I started they could now do easily.

The gap I was building in started closing, and I could see it happening in real time.

Why I stopped

This was the hardest part to admit, but once I looked at it honestly, the reasons were clear.

People already pay for AI. Most people who use AI seriously already have a ChatGPT or Claude subscription. From their side, paying for one more subscription just to merge PDFs makes no sense when the tool they already pay for can do it.

Casual users take the shortest path. Someone who needs to compress a PDF once a month isn't going to search for "Toolchat," upload the file, write a prompt and submit. They'll Google "PDF compressor," click the first link, upload, done. And there are already tons of tools out there doing exactly that.

Power users already have something better. There are people who need this every day, and Toolchat had bulk processing, which was useful for them. But those same people already have a top-tier AI plan. For them it's easier to use what they already have than to depend on another subscription.

The future points somewhere else. I believe more AI progress is coming, and a lot of these features will run right on our own devices. Built in, free, no upload needed.

So I asked myself honestly: would better marketing and more features bring in a large user base? I didn't believe it would. Keeping Toolchat running would just mean burning money on a bet I no longer believed in.

So I closed it. I'm moving on to the next thing.

What Toolchat taught me

An idea that convinces people doesn't always work in the market. Lots of people told me it was a great idea, and I believed them. But "great idea" and "people will pay for it" are very different things.

Retention is everything. Getting people to try something once is the easy part. Getting them to come back is the real game, and that's where Toolchat struggled.

Influencer marketing works. Even on a small budget, it brought real people in. I'll use it again.

Closing a project is part of the journey. It doesn't mean you failed. It means you learned enough to make a decision.

The fun part (because there was one)

I don't want this to sound like a sad story, because it wasn't.

  • At events and networking sessions I got to say, "Yeah, yeah, I'm a founder." That never got old.
  • I always had something real to talk about when someone asked what I was working on.
  • People had questions for me, so conversations actually went somewhere.
  • I got to know AWS bills very personally.
  • And honestly, it was a lot better than doing nothing.

So, was it worth it?

Yes. I built something from scratch, alone: first my own model, then the switch to an LLM, then 30+ tools, then real users from different parts of the world using something I made.

Toolchat didn't make money, but it made me a better builder. And the next one will be better because of it.

If you're holding on to a project you know deep down isn't going anywhere, it's okay to let it go. That's not quitting. It's making room for the next thing.

If you want to see what I built, Toolchat is still up here: toolchat.vercel.app

來源:DEV Community · dev.to