A voice assistant that runs the studio's day

Where it stands

Jarvis is a tool we built for ourselves and use every working day. It runs as a desktop app. A phone line that reaches the same assistant is next. There is no client here and no result to quote. This page describes how the tool is built and the limits built into it.

The problem

Running a small studio means the same ten minutes of checking, over and over. What is on the calendar. What came in overnight. Which leads are worth a call. Whether the websites are getting traffic. None of it is hard. All of it breaks concentration, and most of it happens with your hands full.

What we built

• A window, not a terminal. It opens from the Start menu like any other app. Setup, including connecting Google, happens in a settings panel. Nobody types a command to use it.
• Listening for one phrase. A wake word model listens for “hey Jarvis”, and speech is transcribed on the computer itself. Audio from the room never leaves the machine. Only the words it heard are sent on, once there is something to act on.
• A real conversation. It answers out loud, sentence by sentence as the answer is written, and you can talk over it to interrupt. After it answers, it keeps listening for a few seconds, so a follow-up question does not need the wake word again. In our testing, the first words come back about three seconds after you stop talking.
• Tools, not trivia. It can read and move calendar events, draft email, look up the lead list, read files in Drive, and pull search and analytics numbers for our sites. The first time it opens each day, it gives a short spoken report on how the websites did.
• Work by voice. “Change the pricing page on the studio site” starts a coding task on that project in the background, and the assistant says when it is finished. Publishing only happens when asked for in so many words.

The limits we built in

• Drafts, never sends. Email is written as a draft. A person presses send.
• Asks before it changes anything. Calendar changes are confirmed out loud first, and it never sends invitations unless told to.
• Stays in its own folder. In Drive it can read what it is shown, but it only creates and edits files in a folder of its own.
• Knows what it costs. Every background task has a spending cap, and the day’s running cost is on screen against a daily budget.

Why it matters for your business

The same build fits a front desk or an owner’s day: an assistant that answers the common questions, checks the schedule, and hands anything that matters to a person, in a window your staff already knows how to open. The hard part is not the voice. It is deciding what the assistant is allowed to do on its own, and building those limits in from the first line of code.