MIAOUGPT(1)
NAME
miaougpt — a chat assistant powered by an LLM that runs entirely in your browser
SYNOPSIS
miaougpt miaougpt <model-id>
DESCRIPTION
miaougpt is a simple chat: you type, the (feline) assistant replies, streamed token by token. The model runs fully client-side via WebGPU — no server, no network call for inference.
Model loading, its cache and GPU memory are handled by the central LLM module (see the llm command and the top-right widget, which shows the loaded model and the tokens consumed). By default no model is loaded: on first launch miaougpt proposes a model and asks you to confirm its download.
With no argument, miaougpt proposes a small default chat model. You can also name a specific model (full id or a unique substring); llm --list shows the recommended models.
If a model is already loaded (by llm, glaude or denree), miaougpt reuses it instantly, with no re-download.
CHAT COMMANDS
/exit /quit /bye close the chat /reset /clear forget the context (keep the model) /model show the loaded model /voice [on|off] read replies aloud (Web Speech API) /voice <lang> read aloud in a language, e.g. /voice fr-FR /help list these commands
EXAMPLES
miaougpt miaougpt Qwen2.5-1.5B miaougpt Llama-3.2-3B-Instruct