The first wave of AI character apps did something that deserves credit. It proved that millions of people wanted to talk to fictional characters, not as a novelty but as a daily habit.
Writers tested how an antagonist would justify themselves under pressure. Language learners wanted a patient native speaker who would stay in character through three hundred messages of broken grammar. Roleplayers wanted a scene partner with more time than a human has. The demand was real, and Character.AI, Janitor AI, Chai AI and the long tail that followed built communities in the millions answering it.
But anyone who lived inside those apps for more than a few weeks hit the same four walls.
The first wall: forgetting
You could spend two hours building a character's voice, get them to know that your dog is called Bruno and that you were nervous about your sister's wedding, and come back three days later to a polite stranger wearing the same name.
The model was reconstructing a personality from the most recent stretch of context, and everything older had quietly fallen out the bottom. People worked around it with pinned blocks, lorebooks and ritual re-introductions.
The workaround was the whole problem. A platform that makes you maintain your own memory system has moved its job onto you and called it a feature.
The second wall: the signup
Every character app was its own world — its own account, its own download, its own notifications competing on your home screen.
A friend recommends a character. Now you need to install an app, create an account, verify an email, accept terms, and only then say hello. Most people never got to hello.
The third wall: nothing for the author
Writers who built genuinely good characters — voice, internal logic, the kind of careful psychology that takes hours to dial in — had no way to earn from that work. They gave away craft, and the platform monetised the traffic that craft attracted.
That pattern is familiar from a decade of social platforms, and it ends the same way every time: the best creators stop, or move, or burn out.
The fourth wall: one direction only
A conversation went forward and only forward. If you wanted to see what would have happened had your character said the harder thing two messages back, you deleted forward and hoped the model rebuilt the context you had just destroyed.
Why Telegram, and not another app
The first decision shaped everything after it: no separate app, no signup, no password. Your Telegram account is the account. You open a chat with a bot the way you open a chat with a person, and you are already in.
Someone shares a character with you and the link opens a chat. Not a download page, not a signup wall — a chat. You say hello, the character says hello back, and that is the entire onboarding.
It also meant inheriting things that did not have to be rebuilt: notifications that do not cost a battery percentage an hour, search across your whole history, voice messages, image attachments, and the ability to forward a good reply to a friend.
What had to be rebuilt anyway
The early version ran every character in one chat with the bot. It worked, and it had a ceiling — past two or three characters the chat became a mess and memory got tangled. It was a bridge, not a destination.
The destination arrived with forum topics: each character moved into its own room, with its own history, its own pin strip and its own isolation. The mess vanished, and the architecture finally matched how people were already trying to use it.
Against the other three walls:
- Memory is stored as short readable facts, scoped per character, per topic, per person — and you can read, correct or delete every line of it.
- Any reply forks into a parallel thread and both stay alive, so seeing the other version costs nothing.
- Share links attribute the people who arrive through them, and a portion of what they spend goes back to the author.
What it is not
It is not a companion app. Someone who wants one long-term AI partner is better served by a product designed around exactly that, and there are good ones.
It is not a local tool. If you want to run everything on your own hardware with total control, that category exists and wins on control.
It is character-and-scene-shaped, hosted, and inside a messenger you already have open. That is the whole position, and it is a narrower claim than the category usually makes.