Tab

Tab's pitch is the sentence most listings here paraphrase: an AI you text when you need something done, in Messages, with its own computer and browser behind the thread. Its front page lists the jobs by verb: find and book a flight, call the hotel and move the reservation, turn a PDF into a spreadsheet, order the right replacement part, and it says what happens while it works, a question when something is unclear, short updates, the result back in the same thread, and the job continuing after you leave. When a job needs a phone call it places one and waits on hold. When a job costs money it shows the item, the total and the payee, and pays only once you approve.

It is a private beta: you join a waitlist with a number, and one text finishes signup. No price is published and the terms say the wallet and checkout are limited beta features. The company is Terrasoft, Inc.; no founder is named.

What is published about its security is more than most live listings manage. The privacy policy opens with a short version, and one line of it is the row this site records hardest: passwords and card details never go through Messages, and the AI that plans the work receives an opaque reference rather than the password, which is the surrogate arrangement Muse and asaply describe. The security page goes further than any other page read this month on the mechanics: AES-256-GCM on stored credentials inside a boundary the planning model cannot read, card entry through a ten-minute private link into a separate vault that returns only a reference and the last four digits, inputs masked before screenshots, write actions denied by default, and a rule that an unconfirmed outcome is never treated as a success or repeated. What it does not say: whether conversations train a model, which model provider runs it, and who carries the messages.