User guide
The help wiki that ships with the app, published here so you can read it before installing anything. Inside Quartermaster the same articles are one click away under Help, and the playground assistant searches them to answer questions about the app itself.
Looking for how Quartermaster works inside? That lives beside the code it describes: see the subsystem table in CLAUDE.md.
Getting started
What is Quartermaster
Quartermaster is an all in one local inference engine a front end over llama.cpp (text/vision models) and…
Updating Quartermaster
Quartermaster can update itself (Windows release builds only). When a newer version is available, the Sidebar…
Models & config
Loading and swapping models
You don't manually start models most of the time sending a request to a model (from the playground or an API…
Per-model configuration (the cogwheel editor)
On the Models page, the cogwheel opens the per model config editor. It shows the exact launch command and…
Config variants (normal vs fleet-wide)
A variant is a named alternate profile of a model most often a high or low context version. Each variant…
Automatic config generation
Rather than hand writing a config per model, Quartermaster auto generates one at startup: it scans your model…
Backends (install, update, pick an engine)
Settings Backends is where the inference engines live. It has two layers: the managed catalog on top (install…
Multiple ports & per-port model catalogs
Quartermaster can expose several listen ports from one process, each showing a different subset of models…
Playground
Chat playground
The Chat tab is a full chat client for your local models: Reasoning levels : the Configs popover has a…
Playground accounts
The playground runs on its own port ( playground port ) and is gated behind a per user account, separate from…
Web search
Chat can let the model search the web for fresh or niche facts. Search is an ordered failover chain , not a…
Video & audio transcripts (YouTube and beyond)
Paste a link to a video, talk, stream or podcast episode in Chat and the model can read its captions and…
Quartermaster tools (assistant self-service)
Quartermaster tools let the chat assistant look at and, with your approval, tune the Quartermaster instance…
Image generation
The Images tab drives stable diffusion.cpp models as a thread: every prompt adds a turn, and you can keep…
Upscaling images
Quartermaster upscales any image with an ESRGAN/Real ESRGAN model (e.g. 4x UltraSharp) through the realesrgan…
Image segmentation (SAM)
SAM (Segment Anything) models run through the sam3 server backend and turn an image plus a prompt into object…
Speech and transcription
Two playground tabs cover audio. Speech (TTS) is a small studio: type text on the left, pick a voice, and…
Rerank and embeddings
Rerankers and embedding models have no playground UI they are API only, but they load, swap and share the…
Playground settings
Playground Settings (gear in the side rail) has four sections, all saved to your account: General Theme :…
Monitoring & VRAM
Observe: activity, logs, performance, KV cache
The Observe page has four tabs, plus a shared time window (5 min / 15 min / 1 hr / All) that filters Activity…
GPU memory / VRAM
Quartermaster tracks GPU memory live; the gauge in the status rail breaks the used VRAM into labelled…
KV-cache disk save (survive eviction)
KV cache disk save (dashboard Settings → KV Cache) persists a conversation's KV cache to disk so a long…
API & access
API keys and access
Quartermaster serves an OpenAI compatible API , so existing clients work by pointing at your server's URL:…
Tools API (search & YouTube for your own apps)
Besides chat, Quartermaster exposes the same tool executors the playground uses web search and YouTube as…