"Remembers everything about you" shows up on almost every pricing page in this space. No app pulls that off. The best ones get close by stacking three separate systems, and each fails differently.
System one: the context window
This is everything the model can see right now. Your latest messages sit here word for word, and the space is limited. Once the chat outgrows it, the oldest turns fall off.
How it breaks: without a sound. You get no alert and no warning mode. The character just stops mentioning something you covered an hour ago, and it will happily make up a believable reply rather than admit it lost the thread.
System two: stored facts
The app keeps a short list about you: your name, your job, your dog, the fact that you loathe mornings. That list rides along with every message, which is why these details last while richer memories fade.
How it breaks: it fills up, and it hardens. The list has a size cap, so new entries shove out old ones. Worse, a fact saved incorrectly gets repeated every turn until you fix it at the source. Apps that let you view this list, through a memory screen or a notes panel, are much easier to live with than apps that hide it. Kupid AI lists memory across sessions as a feature. Whether you can see that memory is the question to ask before you subscribe.
System three: rolling summaries
Once the history gets long, the app squeezes it down: one paragraph covering the last fifty messages, then one paragraph covering those paragraphs.
How it breaks: it loses detail, and it cannot get it back. Every squeeze throws something away for good. This is what actually lies behind the common complaint that a companion "changed" after a few weeks. Nothing vanished. The history was boiled down to something blander, and that blander version is what the character is now made from.
What to look for when you choose
Two tests matter, and neither shows on a landing page.
Can you see it? A memory or notes screen you can edit turns a black box into something fixable. That one feature splits apps you can live with for months from apps that slowly drift.
Does it survive time away? Chat daily for a week, go quiet for four days, then come back and mention something specific from day two. Apps built around continuity handle this, which is why Nomi earns its conversation score in our ranking. Apps built around scenes do not. That is no flaw, just a different product.
Two habits that work anywhere
Spell out what matters. "Remember that my sister's name is Ana" has a far better chance of being saved than a passing mention buried in a long paragraph.
And re-anchor after a gap. A single line of context at the start of a session costs nothing and puts the facts you care about back in the window, where the model can read them.
Why the limits go unadvertised
The limits are built in. They are not a bug waiting for the next update. Every app here runs into the same wall, and what separates them is how gracefully they cope. You only notice that after two weeks of daily use, which is exactly why we test the way we do.
