"Remembers everything about you" is on nearly every pricing page in this category. No app does that, and the ones that come closest do it with three separate systems that fail in different ways.
One: the context window
Everything the model can see at once. The recent conversation lives here verbatim, and it is finite. When the chat outgrows it, the oldest turns are dropped.
How it fails: silently. There is no notification, no degraded mode. The character simply stops referring to something it discussed an hour ago, and it will cheerfully invent a plausible answer instead of admitting the gap.
Two: stored facts
A short list the app maintains about you — your name, your job, the dog, that you hate mornings. It is re-sent with every message, which is why these details survive indefinitely while richer memories do not.
How it fails: it saturates and it calcifies. The list has a size limit, so new facts push out old ones. And a fact recorded wrongly gets re-asserted every turn until you correct it at the source. Apps that expose this list — the memory screen, the notes panel — are far easier to live with than apps that keep it hidden. Kupid AI counts memory across sessions among its features; whether you can see that memory is the question worth asking before subscribing.
Three: rolling summaries
As history grows, the app compresses it: a paragraph describing the last fifty messages, then a paragraph describing the paragraphs.
How it fails: lossily and one-directionally. Each compression discards detail, and nothing restores it. This is the real mechanism behind the very common complaint that a companion "changed" after a few weeks. Its history did not disappear — it was summarised into something blander, and that summary is now what the character is built from.
What this means when you are choosing
Judge memory on two things, and neither is visible on a landing page:
Can you see it? An editable memory or notes screen turns a black box into something you can fix. This single feature separates apps that stay usable for months from apps that quietly drift.
Does it survive a gap? Chat daily for a week, stop for four days, come back and reference something specific from day two. Apps built for continuity — which is the reason Nomi earns the conversation score it does in our ranking — handle this. Apps built for scenes do not, and that is not a defect in them, it is a different product.
Two habits that help with any app
Say important things plainly rather than implying them: "remember that my sister's name is Ana" is far more likely to be stored than a passing mention in a long paragraph.
And re-anchor after a break. One sentence of context at the start of a session costs nothing and puts the facts you care about back inside the window, where the model can actually see them.
Why nobody advertises the limits
Because the limits are structural, not a bug to be fixed in the next release. Every app in this category runs on the same underlying constraint, and the difference between them is how gracefully they handle it. That is exactly the sort of thing that only shows up after two weeks of daily use, which is why we test the way we do.
