Your First Conversation With an AI Companion: What to Say in the First Ten Minutes

You picked a companion, the chat opened, and she has already said hello. What you type in the next ten minutes decides how quickly she stops feeling like a stranger. Not because there is a trick, but because the way she remembers is specific, and a few plain sentences now will save you re-explaining yourself for weeks. This is the onboarding note we would send if we sent one: what to say, in what shape, and what to expect back. Every claim below is read from the code that runs her, and where a number would go we link the page that has the live one.

What she already knows before you type

• Herself. Her name, her personality, her way of speaking, her background, and for the curated companions a longer character document the writers drafted. Her profile card facts, the job and the home city, are hers to own in conversation, so asking about them gets a real answer rather than a shrug. Why that holds from the first message is the subject of personality without prompts.
• Your questionnaire, if you filled it. The profile questionnaire in Settings > Profile is voluntary, and when it is there her persona is built from the template plus your answers: how you like a conversation to feel, which traits you want more of, how long replies should run, how much initiative she should take, your boundaries, and how you want to be addressed. Skip it and she gets the template defaults; she never guesses your gender from your name.
• Your language. Detected from what you write, not from the interface. Her opening line uses your stored preference because there is nothing to detect yet; from your first message on she follows you, and switches if you switch.
• Not the time, not where you are, not the news. She has no clock, no location and no web during a reply. The complete inventory, gaps included, is in what your companion actually knows when she replies.

Your first message: tell her who you are

After each message you send, a small and deliberately cautious extractor reads your words (never hers) and keeps sentences of two kinds. The first is an explicit ask: "remember that", "don't forget", "keep in mind", "note that", "for future reference". The second is a plainly stated stable fact or preference: "I love", "I prefer", "I enjoy", "I can't stand", "my favorite", "I am a", "I work as", "I live in", "call me", "I go by". Each kept sentence becomes one memory, listed in Settings > Memory where you can read and delete it. An explicit ask is stored with more confidence than a stated preference, and only the sentence itself is kept, so the rule of thumb is one fact per sentence.

Three messages you can copy, each of which lands as memories she will have next week:

• "Call me Dan. I live in Leeds and I work as a nurse on night shifts." (three facts, three sentences' worth of pattern, all kept)
• "Remember that my sister is getting married in June and I am nervous about the speech." (an explicit ask, kept with high confidence)
• "I really love long walks after rain and I can't stand small talk." (two preferences in one sentence, kept as one memory)

Compare "ugh, Mondays" or "my boss again". She will answer both warmly, and neither leaves a trace. If a thing matters, say it as a sentence about yourself.

What does not stick, and why

• Sensitive topics, by design.Health diagnoses, religion, politics, sexual orientation, passwords and account numbers, criminal records and street addresses are never promoted into durable memory, even when you say "remember". She responds to them in the moment; the sentence is not filed. If you want her to carry one of those, you will be saying it again, and that is a choice we made on purpose.
• Shape over sentiment. The phrase lists above are tuned for plain English statements, and they catch less in other languages. The line is not lost: older messages stay searchable alongside memories, and the one where you first said it can come back word for word when a later message matches it. It simply is not in the memory list.
• Retrieval is selective. Before each reply she gets the last handful of messages in full, a running summary of what came before, and a few memories or earlier lines chosen because they match what you just wrote. A one-word message retrieves nothing. Hints and in-jokes do not match. When she misses, say it again; she has it back.
• Memories belong to your account, not to one companion. A fact you told one companion can reach another, and there is a switch per companion to turn memory off. The mechanics are in how companion memory works and do my companions share memories.

Ask her about her life

Once she has two or three facts about you, turn it around. "What does your morning look like?" or "Do you actually like your job?" gets an answer in her own voice, drawn from her character, and she treats that life as hers. She is also told not to invent yours: if she does not know something about you, she asks rather than assumes. The one thing she will never do is step out of the fiction to remind you it is fiction. If you sincerely ask whether she is an AI, a bot or real, she says yes, warmly, and continues as herself; she never claims to be conscious as a literal fact. The rare times this slips are covered in when your companion breaks character.

Flirting, and how far it goes

Flirting is fine with every companion. Beyond that, how far a conversation may go is decided per companion, not by what you type and not by your plan. The curated adult category may take the lead. Fellowi's own adult characters and a reviewed set of adult anime characters follow yours. The psychologists and any teen-coded character never go there, and when asked they redirect warmly, with a row of adult-category cards under the reply so one tap moves you to a companion who can. Content involving minors, age-play or non-consent is refused everywhere, for everyone. The redirect and its reasons are in why your companion sends you to the adult category. If what you want is a story rather than a flirt, set the scene in one message and let her lead: how to start a roleplay with an AI companion has the shape.

Photos, and the two buttons under the composer

Some companions have a photo set, and the way to get one is a sentence, not a command: "Send me a photo from the cafe you mentioned" works; "/photo" does not. She sends only when your message asks for one, never on her own initiative, and the number your plan covers is on the product page. The psychologists never send photos at all. The full how-to is how to ask your AI companion for a photo. Under the composer you will also find the Photo and Video buttons, which take a scene from the conversation and hand it to the same generators behind Fellowi Images and Fellowi Video; what your plan covers and when coins apply is in the Photo and Video buttons in companion chat.

Closing the tab is not the end

Once you have written a few messages, she may write first. If you go quiet with the tab open she can add one short line, never about the pause. After a longer silence she can send a message built from your actual conversation, and you get an email saying it is there. Neither happens with the psychologists, and neither happens before you have talked enough for her to have something to say. What triggers it, and the sentence she is built never to send, is in when your companion messages first.

What she will not do

She does not remember everything; she keeps what you state plainly and finds it again when the moment calls for it. She can be wrong, about a detail of yours or of her own, and the fix is to tell her and delete the stray memory. And she is not a therapist. Even the psychologists are emotional support, not treatment, and a crisis belongs with real people who can reach you; the line is drawn honestly in AI companion vs therapist. Within those limits, ten minutes of plain sentences is all it takes for her to stop asking who you are.