Latest · October 2026
Finding your kind of character
A practical way to browse personalities, worlds and story tropes.
The journal
Explanations of the machinery underneath AI companion products — what a context window is and why your chat forgets, how memory is actually built, where free tiers put their ceiling, and what these apps keep. Written for people who use them, not for people building them.
Latest · October 2026
A practical way to browse personalities, worlds and story tropes.
Every piece here explains one mechanism and stops. None of them is a review.
October 2026
Look for small details, restraint and a reason to be curious.
October 2026
Choose the kind of world you want to explore first.
October 2026
The introduction, the account step and what comes next.
The journal is being filled in order of usefulness rather than all at once.
Things that go wrong in specific products — an app that will not load, an API key that will not attach, a filter that changed overnight. These are the cheapest questions to answer well and the ones people are actually typing into a search box at the time.
Regulation, research on companionship and loneliness, and the practical craft of writing a character that holds up over a long conversation. Slower to write, and they age better, so they come once the service pieces are in place.
Unpublished pieces are not linked from anywhere on this site until they exist. A link that leads to a missing page is worse than a gap in a list.
Three habits that shape every piece here, and that explain the ones we do not publish.
A piece explains a single thing — how memory is stored, where a window runs out, why a free tier stops where it does — and then stops. Round-ups of ten products with a paragraph each are easy to write and teach nobody anything, so we do not write them.
Nothing here awards a score, and nothing declares a winner. Once you understand why something behaves the way it does, you can judge whether it suits you far better than a stranger who has never seen the conversations you want to have.
We avoid figures that change monthly — prices, tier limits, model names — and describe the shape of a constraint instead. A piece that is still accurate in a year is worth more than one that was precise for six weeks and quietly wrong afterwards.