Quartermaster 1.7 adds support for llama.cpp decision models via /v1/systemone

Quartermaster 1.7, a local inference manager for swapping llama.cpp, image, audio, and video models, now supports decision models via the new /v1/systemone endpoint.

Key Features

• Auto-discovery: Dropping a decision GGUF (e.g., Kev, OpenJev, Laya) into the models folder automatically creates a config entry sized for VRAM.
• Playground tab: Users can write prompts, select answer shapes, and view probability bars for each option to quickly test model outputs.
• Model browser: A new Decision tab allows users to find and proxy decision models directly from Hugging Face.
• API integration: POST /v1/systemone requests go through Quartermaster's scheduler and API keys, allowing decision models to swap in alongside chat models.

Requirements

• Requires a llama.cpp build new enough to support /v1/systemone (upstream b11361+). Quartermaster will warn if an older build is detected.