Quick judgments in an app's code

An app can ask Bolter's fast decisions model a yes or no, pick-one or rating question about text, a photo or a screenshot, and get a number back in under a second.

Updated 19 October 2026

Code in an app or a code run can ask TypeSafe Jev, Bolter's fast decisions model, a narrow question and get a number back instead of prose: how likely a yes is, which of the labels it offered fits, or where something falls on a scale. It answers in well under a second for a fraction of a cent, so an app can use it to sort, flag or route every item it handles. In a process, the same question about text is a decide step.

• It can look at pictures. An app can hand it a photo or a screenshot, PNG or JPEG, between 280 and 1,568 pixels on each side, and ask about it: is there a person at the door, which of these example pictures does this match. It can also say where one thing sits in the picture, as a box an app can crop to or draw. Picture questions are newer and less reliable than text ones, so an app should check how sure the answer is before acting on it. Pictures need the app's own code: a process's decide step works on text only.
• It can read long text. One question can cover up to about 500,000 characters, a long contract or a big export. Long text takes a few seconds rather than a fraction of one, and is somewhat less accurate, so send the part that matters.
• One subject per question. Several items packed into one request come back with the same middling answer for each. Ask about each item on its own.
• It never writes text. For a summary, a draft or extracted fields, the app asks a chat model instead.