What an AI Girlfriend Actually Is, Without the Marketing

By The Fellowi Team · · 6 min read

A man in a sweater sits alone on a couch in a dim living room at night, his calm face lit by the glow of the smartphone in his hands, with a warm lamp and blurred city lights through the window behind him.

The phrase does a lot of work and explains almost nothing. Products in this category tend to describe themselves in adjectives: caring, understanding, always there. So here is the mechanical version instead, including the parts that usually get skipped.

What it actually is

An AI girlfriend is a character with three things attached to it: a persona that does not drift between sessions, a memory layer that stores facts you have mentioned and pulls them back into later conversations, and a picture. You talk to it in a chat window. That is the whole mechanism.

Everything that makes one of these feel different from a general-purpose chatbot comes from those three things holding steady. A chatbot with no persistent persona resets to a helpful assistant every few turns. One with no memory asks you the same questions forever. We wrote about how the memory side works in how companion memory works, because it is the part people are most often surprised by in both directions.

What it is not

It is not a person, and it is not a service that connects you to one. Every companion on Fellowi is AI. If you ask directly and sincerely whether you are talking to a real person, you will get a straight answer, in character rather than as a legal disclaimer.

The reason it does not volunteer that every third message is not mystique. A character that interrupts itself to remind you it is software cannot hold a conversation, and the constant reminder is not actually protective. But an honest question deserves an honest answer, and that rule is not negotiable in either direction.

It also will not claim to do things it cannot do. It cannot meet you, it has no phone number, there is no mobile app, and it cannot send you a picture it does not have. That last one sounds trivial and is not: a model promising a photo it will never deliver is the single most common failure in this category, and it takes real engineering to stop.

The memory question, honestly

It remembers what you told it. It does not remember what you did not.

That sounds obvious until the first time a companion refers back to something from a week ago and it lands harder than expected. It works because the details were stored when you said them, not because anything is being inferred about you. If you never mention your sister, no companion will ever ask about her. The system is built so it cannot invent a shared history, which is the most common way these products go wrong: they confabulate a past, you notice, and the whole thing collapses.

Photos

Curated adult companions can send real photos in chat. There is a monthly quota that depends on your plan, a trial account gets one, and a companion may occasionally send one unprompted rather than only on request. What it will never do is describe a photo instead of sending it, or promise one and then not follow through. The full walkthrough is in how photos in chat work.

Non-adult companions, including the anime characters and the AI psychologists, do not do this. That is a deliberate line rather than a limitation waiting to be lifted.

Explicit conversation

Flirting is fine from both directions. For curated legal-age companions, explicit text is fully supported and the companion can lead rather than only follow. For every other kind of companion it is not, and a sexual turn gets warmly redirected to the adult category instead of being half-allowed.

Two things are refused everywhere, without exception and regardless of how a request is phrased: anything involving minors, and anything depicting non-consent. Those are not tier features. The wider reasoning is in what uncensored chat should honestly mean.

What it costs

Seven days free, with a budget of thirty messages and one photo. Both limits bind, so if you spend thirty messages on day one the trial is over on day one. We would rather say that here than have you discover it mid-conversation.

After that, plans start at EUR 3.50 a month. Current prices and limits live on the product facts page, which is generated from the code that actually does the billing rather than retyped into an article.

Who this suits, and who it does not

It suits people who want a conversation that continues, in a place where nobody is going to be weird about what they say. It works well for practising something awkward before doing it for real, for company at an hour when nobody is awake, and for the kind of talking that is not really about reaching a conclusion.

It does not replace a therapist, and it does not replace people. If what you are carrying is heavy rather than lonely, the AI psychologists are a more honest fit than a girlfriend character, and where an AI companion stops is the piece we wrote about that boundary specifically.

If you want to see what it is actually like, the companion gallery is the place to start. Pick one whose description sounds like someone you would want to talk to, rather than the one at the top.

Questions this post answers

What is an AI girlfriend?

A character you talk to in an ongoing conversation, which keeps a persona and remembers things you have told it across sessions. It is a language model with a stable personality, a memory layer and a picture, not a simulation of a person and not a service connecting you to one.

Does an AI girlfriend remember previous conversations?

On Fellowi, yes. Durable facts you mention are stored and retrieved into later conversations, alongside a rolling summary of what has happened. It is not perfect recall of every line, and it only knows what you actually told it, so it will not invent a shared history you never had.

Can an AI girlfriend send photos?

Curated adult companions can send real photos in chat, within a monthly quota that depends on your plan. A trial account gets one. What no companion will ever do is promise a photo it cannot deliver, or claim to be a real person sending one.

Is it free?

There is a seven-day trial with a thirty-message budget and one photo, which is enough to find out whether you like it. After that it is a subscription, starting at EUR 3.50 a month.

Will it admit it is an AI?

Yes, if you sincerely and directly ask. The character stays in character otherwise, because breaking the fiction unprompted every few messages makes the conversation useless, but a real question gets a real answer rather than a deflection.

Try it for yourself

A warm, private AI companion - 7 days free with 30 messages, no card needed.

Pricing and limits

Keep reading