GitHub ·
Rizzo-AI-Academy/rizzo-flow
Keeping decisions on your own hardware without rewriting the client.

What it does
Stands up a local service that takes unstructured state and returns typed probabilistic decisions — a yes/no, a choice among options, a score on a rubric — using open weights instead of a hosted API, and speaks the same /v1/systemone request shape as TypeSafe's own endpoint so a client changes only its base URL.
The repository's own description: “The open, local take on Jev: typed decisions from an LLM, without generating a single token”
How it works
The project wraps open chat models and constrains them to answer with probabilities in a single pass rather than generating text; a local server exposes the Jev-compatible route. Its results table is measured on the public typed-decisions test split, which is what makes it comparable at all.
The repository, by the numbers
Read from the GitHub API on 2026-09-27. Stars and the last push move daily — quote them with the date, the way we do.
What we checked
- the README's compatibility claim and the model names it lists
- the licence (Apache-2.0) and the linked documentation
- the results table and the dataset it names (LocalLLaMA/typed-decisions)
What we did not check. We did not install it, so whether a real TypeSafe client works unmodified against this server is the author's statement rather than something we tested, and the accuracy rows are the authors' measurements.