Something happened and nobody explained it.
The character forgot a thing it had been told three times. It contradicted a detail from last week. One Tuesday it started answering differently — same name, same description, noticeably not the same. A conversation that had been going fine hit a message cap. A feature you were using stopped existing after an update.
Every one of those has a boring, mechanical, entirely knowable cause, and none of the available writing about companion apps will tell you what it is. The listicles are selling you something. The think-pieces are talking about you rather than to you. Both skip the part where somebody explains how the product works.
That is all this site does.
What you are talking to is a language model, a written description of a character, and some portion of your conversation history — reassembled from nothing every time you send a message. There is no persistent entity sitting there between your messages. Once that is clear, most of the mysteries stop being mysterious. Memory is a feature somebody implemented and it loses things because every implementation of it loses things. A character changed because a company changed a model, edited a description, or adjusted a filter. The cap exists because every message costs the operator money.
The theme running through all of it: the operator decides. Not the character, and not you. Which model, how much history it can see, how long messages are kept, what the filters allow this month, what a message costs. Every post here ends up locating a decision, and the useful thing about knowing where the decisions live is that you stop attributing them to the character.
Some ground rules, so you know what you are getting. No app is named anywhere on this site — naming one would be an endorsement and any claim about a specific product’s current behaviour would be out of date within months. Nothing here reviews, ranks or recommends. Nothing here has an opinion about whether you should be using one of these apps; that is your business and it is not the subject. And there is nothing explicit on this site: it is a consumer explainer and it reads like one.
What you get instead is the mechanism, who chose it, and what you can actually check — including the specific wording worth looking for in a privacy policy, which is usually the only document that answers the questions people most want answered.
Latest posts
-
Nothing is there between your messages
A companion is assembled from scratch on every turn: a model, a written character description, and whatever history fits. Nothing persists in between.
-
What memory actually means
Memory is a feature somebody built, not a property of the model. Every way of building it loses something, and the losses have recognisable shapes.
-
Why the persona changed
Same name, same description, different behaviour. Five mundane causes — and all five are decisions made by a company on a date, not something that happened.
-
What happens to your messages
Your conversation is rows in someone else's database. Retention, human review, training use and deletion are all answered by documents you can read.
-
How the pricing works
Every message costs the operator money, which is why caps, credits and short memory exist. The economics explain most of the product design.