Local Models

Run open models entirely on your own computer from the Hermes desktop app: no account, no API key, and no context sizes, GPU layers or quantization to tune.

Local Models is on by default in the Hermes desktop app on macOS, Windows and Linux. Hermes downloads and manages the llama.cpp engine, picks the build of each model that fits your hardware, and manages memory end to end. Open Settings, Providers, Local Models, click Install runtime, download a model from the catalog, and click Use; new chats then run locally, and going back to a cloud provider is one click in the model picker. From a terminal, hermes model then Local models runs the same setup. Every catalog row is priced against your machine: green fits your GPU, amber uses system RAM and runs slower, red is too big. Hermes never offers builds below 4-bit, starts each model at a context window that fits and grows it as the chat needs room, gives every recommended model at least a 64K window, and unloads idle models after 15 minutes. Find more models searches all of Hugging Face, Add model file links a .gguf you already have without copying it, and a llama-server you already run is detected and used. Once a model is downloaded, nothing leaves your computer.