Why I stopped billing my users for AI tokens

The AI part of my SaaS was 10% of the product and 80% of my support tickets.

"Why did my bill spike?" "I only used it three times." "Is there a bug?"

No bug. Token counter.

I'd done what most devs do — grabbed the OpenAI API and passed the pricing model straight through to my users, which seemed reasonable. But after all users were frustrated and confused by increased bills. Even more painful was their unpredictability.

I was full-stack developer with no particular interest in AI/ML. And what I understood, is that I just want to send API request and get a classification, extraction, translation, etc. — the stuff that's in every app and never gets a blog post.

I dived into AI and ML for several months and finally started building new service: a8k.me. Simple, flat monthly rate, free tier and 100% private - inputs never saved or logged.

Of course there is a tradeoff: you give up control. No picking models per task, no prompt engineering for token efficiency. If model selection is core to what you're building, this isn't for you.

But if AI is a feature in your product and not the product — and you're spending brainpower on billing logic and support tickets instead of the actual thing — it might be something you were looking for.

It's still on early stage, only one mid-size LLM model is available. But we are working on Speech-To-Text, OCR and other models implementation.

What do you think? Is this a step in right direction?