Show HN: Jevper – the Jev interface on top of any OpenAI-compatible model

Jevper is an open-source implementation of the Jev interface that works with any OpenAI-compatible language model, allowing developers to get typed responses (choices, questions, scores) with probabilities and confidence scores. Unlike the proprietary TypeSafe API, it can run on self-hosted models like llama.cpp, enabling the same code to work across different backends without modification. The tool uses pydantic as its only runtime dependency and supports multiple inference methods including logprobs, grammar constraints, and structured JSON schemas.