An AI glossary, translated into what kisir actually does

Alphabetical, as a glossary should be. Each entry has the word, the usual meaning on an app page, and the translation into what kisir actually does. Several translations are the same sentence, which is sort of the point.

Assistant. An AI you talk to inside the app. In kisir: the Add Expense button. It has never asked how your day was.

Automation. The app does things on your behalf. In kisir: when every balance in a group reaches zero, the group closes itself. If someone adds an expense later, it asks before re-opening. That is the full list of things it does on its own.

Categorization, automatic. A model sorts your expenses into categories. In kisir: a description field. “Fuel” is filed under “fuel”.

Confidence score. How sure the model is. In kisir: not a feature, because €127.50 divided by six is €21.25 and there is nothing to be unsure about.

Context. What the model knows about your situation. In kisir: the group. Who is in it, what was paid, who it was for. The app knows nothing outside the group, and nothing about you outside the name you typed.

Fine-tuned. Adjusted to your data. In kisir: the exchange rate. The app fills in today’s rate, and you can fine-tune it to what your bank actually charged. That is the only thing in the app that is tuned, and you are the one tuning it.

Hallucination. The model states something that isn’t true, confidently. In kisir: when your friend is sure they paid for the taxi. The expense history (who changed what, old number to new) is the cure. The app itself can’t hallucinate. It can only add.

Insights. Charts and observations about your spending. In kisir: the Overview screen. Your balance, everyone’s balance, who spent how much. For actual charts, export the ledger to Excel and make the ones you want.

Learning. The app gets better the more you use it. In kisir: no. The app on day 300 divides exactly as it did on day one. You get better at it, which is the kind of learning that was always meant.

Memory. The assistant remembers past conversations. In kisir: the ledger. Every expense, every payment, every edit, kept so the group can check it. Nothing about you is remembered anywhere else.

Model. The thing that makes the predictions. In kisir: there isn’t one. There is a function that takes balances and returns the fewest transfers to settle them. It returns the same thing every time for the same input, which disqualifies it from being interesting.

Natural language input. Type a sentence, the app works out the expense. In kisir: a form. The form has the total, who paid, who it was for, and how to split it. It is slower to describe than to fill in.

Nudge. A notification timed for maximum effect. In kisir: a notification when something happened. Someone added an expense, recorded a payment, joined, settled up, or changed an expense you’re on. The title is the group’s name. It is not timed.

On-device. The model runs on your phone, not a server. In kisir: everything runs on your phone, including adding expenses with no signal. They sync later. There was no model to put on a server in the first place.

Personalized. Tailored to you. In kisir: your display name, which can be different in each group. And “Your plan”, your part of the settlement. That is the extent of the tailoring.

Predictive. Guesses what you’ll do next. In kisir: see Confidence score.

Prompt. What you type to the AI. In kisir: the six-character code you type to join a group. It has one correct response, and it’s the group.

Receipt scanning. Photograph a bill, the app reads it. In kisir: not included. Typing the total is faster than checking a scan, and the person who paid has the total in front of them. The longer argument is in why kisir does not scan receipts.

Recommendation. The app suggests what you should do. In kisir: the settlement plan. It recommends that Dan pay Ana €40.00, because Dan owes €40.00 and Ana is owed it. It is the only recommendation the app makes, and the settle-up guide lets you reproduce it by hand.

Smart. Adjective attached to any feature with a model behind it. In kisir: adjective not used. The split is Equal, Shares or Amounts. None of them is smart. All of them add up to the total, or the editor won’t save.

Summary, AI-generated. A paragraph about your trip, written for you. In kisir: the Overview, which is a list of numbers. If you want prose, the trip was the prose.

Training data. What the model learned from. In kisir: nothing was trained on your groups, because nothing was trained.

Trust. A word used in the sentence “we take your trust seriously”. In kisir: the thing the app is for. Four friends believe one list because every number in it can be checked, nobody can quietly change someone else’s expense, and the settlement is the same whoever computes it. That is not a feature. It’s the reason the other entries say “no”.

If you are looking for a group expense tracker and this vocabulary is on every page you open, the trip expense tracker guide explains what the app actually has to do, with or without kisir, and none of it needs a glossary.