Live capture from the AgentRQ app — description in, generated title out, entirely client-side.
Three steps, all of them inside your browser tab.
Once your task description passes five characters, the sparkle button appears next to the title field — it stays hidden the rest of the time so it never gets in your way.
The first time you click the sparkle, your browser
fetches a small model via @huggingface/transformers
and shows real download progress — "Loading AI Model... 42%". After that, it's cached.
Every generation after the first is local inference — no network round-trip. Edit the generated title any time; typing over it simply overrides the suggestion.
Because the model runs in a Web Worker inside your tab, the field shows you exactly what state it's in — Loading AI Model... N% during first-time download, then Generating title... while it's inferring. There's no black box spinner hiding a server call.
Your task description — which may include internal file paths, hostnames, or credentials-adjacent context — never leaves the browser to produce a title suggestion.
