Quartermaster
An all-in-one local inference platform that computes each model's launch flags to fit a user's VRAM, hot-swaps between chat, image and voice models, and serves them behind one OpenAI- and Anthropic-compatible API.
Quartermaster is pointed at a folder of models and works out what fits in the available VRAM, launches each one with computed flags, and swaps between them on demand behind a single OpenAI- and Anthropic-compatible API. It does not run models itself; instead it orchestrates existing inference servers the user already trusts, such as llama-server, stable-diffusion.cpp, whisper.cpp, vLLM and TabbyAPI, and shuts them down when idle.
It ships its own chat application with per-person logins, web search, image generation and voice, reachable from any phone or laptop on the local network, making it usable as a private household ChatGPT alternative. It distributes as a single Go binary configured with one YAML file, grew out of a fork of llama-swap, and sends no telemetry.