Jevper

jevper is an independent, unofficial implementation of the documented Jev wire format, described by the project as System One, layered on top of any client that exposes an OpenAI-compatible responses, chat completions or messages call, importing no provider software development kit itself. A single function call takes a state description and a set of typed questions, choice, score or yes/no, and returns typed answers carrying probabilities and confidence, working the same way whether it talks to a hosted Jev endpoint, a self-hosted llama.cpp server, or another package's own wire format.

Its default method asks for token log-probabilities where the connected provider returns them, and falls back to asking for a JSON answer where it does not, remembering which approach worked for a given model and surface. Each question in a request is sent as its own provider call and questions run concurrently up to a configurable limit, with answers returned in the caller's original order keyed by question id.