What Ambient AI Looks Like on the Desktop.

Jarvis (getjarvis.eu) is a screen-aware desktop AI assistant that stays out of the way until you press a hotkey. Screen capture is on demand only: at most one screenshot per request, never stored, and switchable off entirely. Cmd+/ on macOS or Ctrl+/ on Windows summons a floating bar that sees the screen on demand and can point at what it means. 30+ app connectors reach Gmail, Slack, Notion, and Linear, and persistent memory keeps the thread across sessions.

I started building this product because I saw one obvious gap in the people I knew: AI has gotten super smart and continues to get smarter, and yet most of the people around me never used it more than a few times for conversation. So I wondered why, and I came up with the assumption that if AI was easier to use, more accessible, and more personal, then people would start using it more — and add a lot of extra value to society by doing so.

So I sketched out in my mind how this product should look, and based on my long experience with AI, I noticed two more things:

1) AI was a chat window, or a tab — a separate thing to go to. It wasn't integrated with your computer; it wasn't one click away from asking a question. That level of friction, I thought, might be one of the reasons people haven't truly adopted it to its full potential yet.

2) AI in its current form doesn't have enough data about you. It just doesn't. You want to ask it very personal, tailored questions, but it fails and gives generic responses because it doesn't know nearly as many things about you as you do.

So I decided to do three things — none of them completely new on their own, but the combination of them, I hope, makes them powerful and maybe even revolutionary:

a) I gave Jarvis access to my screen. Whenever you ask a question, Jarvis can see what you see, so it can help you in a ton of situations — guiding you, or simply learning things about you. Of course, there's a small button to turn that off at any time if you want privacy.

b) I gave Jarvis access to all the apps I use. The truth about most people is: they use a lot of apps. And getting AI to know everything about your work is almost impossible, unless you give it the same access you have. Whenever you connect an app, Jarvis goes into it and studies everything in the most minute detail, so it never misses even the smallest things about you. As a bonus, it can perform tasks for you while it's in them. So if you need to message someone, check something for updates, or run a complex multi-app task, this is perfect. It's not on the level of an employee that can do all your work — it's more like a secretary or informant. It can use the apps, but it still leaves the work to you most of the time.

c) I gave it great memory. Jarvis has what I call "human-level" memory. It's divided into long-term and short-term, so information that's only useful for a short period — like a meeting you have with someone — can be forgotten when it's no longer needed. It can also divide things into categories, and it knows the exact time where you are so it's always on time.

All of these things combined have, in my experience and in that of others who've tried it, truly been a huge improvement in the way we use AI. We're slowly moving towards more integrated systems that know more about us and are easier to access — and as AI gets smarter, we too need to find new ways to use it and new tasks to delegate. Increasing our overall contribution to society isn't about fearing AI; it's about embracing it more, in smarter ways.

Vision