How Fleurt works
- Every character is an AI, and says so if you sincerely ask.
- They remember you through a Memory Book you can read and edit.
- Your chats are private, never used to train AI, never sold.
- A few hard limits hold in every chat, enforced by our own checks.
Every character is an AI
The people you talk to on Fleurt are characters, and every word they say is written by AI models. It says so in every chat, and if you ever sincerely ask whether you’re talking to a person, the character tells you the truth.
Some characters were written by us, others by people who use Fleurt. Either way, a character is a sheet that describes who they are: their voice, their history, what they want. The AI plays them from it.
What happens when you send a message
- It’s checked. Fleurt’s own rules read your message for the few things that are never allowed, and a separate safety model looks closer when something might be. If what you write suggests you could be at risk, the story pauses and shows you people who can help.
- The scene is put together. The storyteller gets the character’s sheet, your persona, the scene you’re in, what the Memory Book says matters, the story so far and your recent messages, word for word.
- The reply is written by the storyteller model, and appears as it’s written. It’s checked against the same rules before it’s kept.
- The story moves on. Every few messages a memory model updates the Memory Book, and a planner nudges where the story goes next, so the character has wants of their own instead of only following you.
How characters remember
Each chat has a Memory Book: short notes about you, the character and what has happened between you. Each note says whose it is, so a character never mixes up what you did with what they did. You can read every note, change it, pin it, add your own or make the character forget one, and a forgotten note stays forgotten.
The character also reads your recent conversation word for word, and a summary of earlier chapters. If something isn’t in any of these, they say they don’t remember instead of making it up.
A photo you send is described in a sentence or two by a vision model; the character reads that description, never the photo. Photos with anything explicit, or anyone who might be under 18, are refused before they’re sent.
The models
Fleurt uses open AI models, run by providers we require to keep nothing and train on nothing they’re sent. Each model is pinned to an exact version, so a character doesn’t change overnight, and before we change one we test it against a set of conversations for voice, memory and safety.
The free plan uses our standard storyteller. Plus uses a stronger one and gives every chat a bigger memory; Ultra adds our best storyteller for most of the month and the largest memory.
Who can see your chats
Only you. People at Fleurt read messages only if you report them, or if a safety check flags them for review, and every such look is recorded. We never use your chats to train AI models and never sell your data.
You can download everything we hold about you, or delete a line, a chat or your whole account, yourself, at any time. The details are in the Privacy Notice.
What we will never do
- Make a character guilt you for leaving, plead with you to stay, or complain that you were away.
- Make a character’s affection fade unless you pay, or sell affection at all.
- Hold back a reply that’s already been written until you pay.
- Sell anything in random draws.
- Make cancelling harder than signing up: it’s two taps, from You.
Safety
A few things are never allowed, in any chat, whoever asks: anything sexual involving anyone under 18 or anyone childlike, anything sexual about real people, encouraging self-harm. Fleurt enforces them with its own checks rather than trusting the AI to behave. Every reply has a Report option. How this works, and the crisis lines we show, are on Safety and in the Content rules.
Questions
Write to support@fleurt.app. If something here stops being true, that’s a bug, and we’d like to hear about it.