Developer Releases Von, an Open-Source 395M System One Model Replacing TypeSafe JEV
Developer wfzyx has released Von, an open-source 395-million parameter "System One" model designed as a direct drop-in replacement for TypeSafe's commercial JEV model. Available on GitHub and Hugging Face, Von reportedly outperforms JEV across all benchmarks while delivering sub-second response times. This release provides developers with a free, self-hosted alternative for ultra-fast, structured AI decision-making without relying on proprietary cloud APIs. It highlights the growing trend of ultra-lightweight, task-specific models capable of running local inference on consumer CPUs without needing dedicated GPUs. Von runs entirely on a CPU with only 1–2 GB of RAM and delivers response latencies ranging from 25 to 300 ms. Although designed primarily for low-resource CPU usage, running the model on a GPU further reduces latency.
## BACKGROUND
Inspired by the psychological concept of fast, instinctual thinking, "System One" AI models focus on making instant, structured probabilistic choices rather than generating long text responses. TypeSafe AI popularized this concept with JEV, a proprietary model designed to evaluate program states and return typed answers directly to software systems.