Would You Trust an AI That Never Touches the Cloud?

Every "AI writing assistant" promises privacy… then quietly streams your drafts to someone else's server. RephrAIs flips that script by running entirely offline in your browser, using compact open‑source models like Llama and Phi‑3 accelerated on your local hardware. No prompts, emails, or DMs ever hit a remote API – which is either incredibly reassuring or totally overkill, depending on how paranoid you are about data.

I'm curious where PH builders land on this:

• Is true on‑device AI (no cloud, no logs) a must‑have or a nice‑to‑have?

• If you're a lawyer, doctor, founder, or handle sensitive client work, would you actually switch from Cloud AI to something like a RephrAIs bookmarklet that lives in your browser?

• Do you worry more about model "quality" or about your content being scraped, logged, or used for training?

Also: what's the "oh sh*t" scenario that would finally make you abandon cloud‑based AI tools for a privacy‑first rephraser? I want your spiciest threat models, real horror stories, or reasons you don't care at all.