Decisions your code can branch on.

What Wity is for#

Most of the decisions in a product are small and bounded: which queue gets this ticket, does this message break a policy, is this passage relevant, how severe is this alarm. A chat model can answer them, but you get free text to parse, no honest sense of how sure it is, and latency and cost sized for essays. Wity is built for exactly these decisions.

• Typed answers. You declare the shape (choice, noul or an ordered score) and Wity can only answer inside it. No parsing, no invented labels.
• A probability on every option. Not a vibe or a single "confidence": the full distribution, so your code can act on sure answers and route unsure ones to a person.
• Thinks only when it matters. In auto mode clear questions come back in about a tenth of a second, and Wity reasons step by step only when an answer is genuinely close, then stops as soon as it has settled.
• Documents in, items out. split cuts a thread, contract or log into its emails, clauses or entries in under a tenth of a second, without changing a character, and learns your format from an example. New.
• Fields out of any document. extract returns the values you name, as written in the document, with their position and a confidence. New.
• Honest odds on "how likely" questions. When a question asks for a forecast, Wity works the probabilities out from the evidence instead of guessing a side.

One request, one decision#

A request carries state (the text or JSON the decision is about) and one or more named questions. Here a support message is routed:

Your code branches on answers.intent.choice and can check answers.intent.probabilities before acting on its own.

Where to go next#