Key Info

The latest llama.cpp builds introduce a /v1/systemone endpoint enabling Jev-style decision inference locally, efficiently, and privately across multiple open models.

Highlights

  • New /v1/systemone endpoint available in recent llama.cpp builds
  • Supports Jev-style inference: ranking options, estimating yes/no probabilities, scoring ordered scales
  • Runs fully locally for privacy, with no generated text required
  • Multiple open models supported, with more expected
  • A Gradio playground demo lets users give Julia-1 a situation and watch decisions shift as context changes