Ninja AI — thirteen AI models, one chat

A dozen services
instead of
one chat

To use AI models properly you need accounts on several services, a foreign bank card to pay for each, an understanding of which draws better and which writes better — and you have to keep all of it in your head.

Most people simply don’t need that. They need one chat where they can ask and get an answer. Ninja AI is that chat: no sign-up, and payment right inside Telegram.

What makes it hard

Wiring a chat to an AI model is an evening’s work. Making something people use and pay for is a different job entirely.

Question“Write a launch post and make a picture for it”

→

The bot works outwhat is being asked fortextimagevideovoice

→

Model choicethe one that does it better and cheaperif it is down, the question moves to the next

→

Answerin the same chatcharged once; refunded if it fails

01

The bot makes the choice

The user does not need to know how one model differs from another. The bot works out from the question what is being asked for — text, an image, a calculation or a long analysis — and hands it to the model that does it better and cheaper. You can also pick one by hand if you like.

02

Failures must stay invisible

AI services go down and slow down regularly. When that happens the question quietly moves to the next model, and any provider that fails three times in a row is temporarily dropped from the rotation. The person in the chat sees nothing but the answer.

03

The money has to add up

A user notices an extra charge faster than a poor answer. So a payment cannot go through twice even if the system receives it again, and a video that fails to generate is refunded to the balance automatically.

What happens when a model fails to answer

→ the diagram scrolls sideways

The outcome

A finished service: subscriptions, payments, support, bonuses and an admin panel, all inside the messenger people already use every day.

13 AI models answer as one — no switching between services

0 websites or sign-ups: payment and account live right in the chat

1 subscription instead of a dozen services — paid for right inside Telegram

Built with: Python, aiogram, PostgreSQL, Redis, Docker.
17,714 lines of code · ran with real users, now handed over to the client with launch documentation · my role: architecture, development, launch.