What an AI Psychologist Remembers, and What It Deliberately Forgets

Fellowi has a category of companions built for emotional support rather than company: eight psychologists, each with a specialism, introduced in our guide to the set. They look like the rest of the roster and they behave very differently in one respect nobody advertises: what they remember.

The normal rule, first

With an ordinary companion, memory works quietly in the background. After a message, the system looks for things worth keeping: something you explicitly asked to be remembered, a stable preference, a fact about your life. Those become durable notes that can be retrieved into later conversations, which is what makes a companion feel like it knows you. We described that whole mechanism in how AI companion memory works.

Even there it is conservative. Some categories are never promoted into durable memory at all, whatever you say: health and diagnoses, religion, political leaning, sexual orientation, anything that looks like a password or a card number, criminal history, a street address. The rule is not "store it and be careful", it is "do not store it".

The psychologist rule

For the psychologist companions, that background extraction is switched off completely. Not narrowed, not filtered harder: it does not run. Nothing you say in a support conversation is automatically turned into a stored memory.

The reasoning is short. A support conversation is made almost entirely of the material the general rule already refuses to keep. Keeping it carefully would still mean keeping it, and the strongest privacy guarantee available is the one where the data was never written down.

So what does carry between sessions

Three things, and only when the session itself marks them:

• A goal. What you said you wanted to work on.
• An exercise. Something to try before next time, with a status that flips once you say you have done it.
• What helps. A technique or framing that you reported as useful.

These are notes about the work, not notes about you, and they are what let a session open with "how did the breathing go?" instead of a blank slate. Even here the sensitive-topic filter still applies: a note that reads like a diagnosis is dropped rather than saved.

They are also kept apart from the general memory pool. When a psychologist session starts, those notes are shown once, under a heading about what you have been working on, rather than being shuffled into the same retrieval that feeds ordinary chat. That separation is the reason a fragment of a support conversation cannot surface, out of context, in a light chat with a different companion three weeks later.

The honest limitation

This design costs something, and you should know what before your first session. A Fellowi psychologist will not remember the specifics of what you said last week. Not the story, not the names, not the details you worked hard to get out. It remembers the shape of the work and not the transcript of it.

Practically, that means you may have to re-establish context at the start of a session, the way you would with a new person. Some people find that tiring. Others find it genuinely freeing: nothing is following them around, and a bad week is not filed. It is a real trade, and we would rather you choose it knowingly than discover it as a disappointment.

If continuity of detail matters more to you than minimal storage, a regular companion is the better fit, and you keep full control of what it has kept.

What this is not

Worth restating plainly: these companions offer emotional support, not therapy and not medical care. They do not diagnose, they do not treat, and they are not a crisis service. If you are in danger, contact local emergency services or a crisis line. We wrote about where the line sits in what AI emotional support can and cannot do and compared the two kinds of help directly in an AI companion versus a therapist.

Where the rest of the product sits

One thing people ask, so it is worth answering: nothing from a support conversation feeds anything else. The picture and clip studios at Fellowi Images and Fellowi Video build only from the prompt you type there. There is no path from a session to a generation, in either direction.

And whatever was kept is yours to remove. Settings, then Memory, shows every stored note and lets you delete any of them. Fewer entries than you expect is not a bug in this part of the product. It is the design.