Saint wants to be your computer operator, but it's not available yet
Sh0ny
6 августа 20262 min read
In short
Saint promises not only to respond to requests, but also to view the screen, control apps, and remember context across sessions. But the main issue here isn’t the convenience of the interface; it’s how safe it is to hand over control to an agent—especially since the product is still in development.
A personal AI agent that can see your screen and click on apps on its own could significantly change the way you work on a computer. But for now, Saint isn’t a finished tool—it’s a promise: the product is in active development, available only through a waitlist, and the price hasn’t been set yet.
The idea is straightforward: instead of a long prompt, you can simply say “organize this table,” “book a ticket,” or “explain this homework page.” Saint is designed to read the screen, search for up-to-date information online if necessary, and then carry out the task in Chrome, VS Code, Slack, Gmail, Notion, Excel, Figma, Terminal, or YouTube.
The strongest aspect of the concept is its cross-session memory. The agent promises to save projects, preferences, and past decisions so that every conversation doesn’t start from scratch. For studying, this could mean continuing work on a research project, and for everyday tasks, it means less repetitive data entry.
But this memory also transforms Saint from “just another chatbot” into a trusted computer operator. It potentially gains access to your messages, documents, email, and financially significant actions. Therefore, more important than an impressive list of integrations is exactly how the agent limits its permissions and how often it makes mistakes.
The limitations are still significant: Saint hasn’t been released yet, support for Windows and macOS is only planned for launch, and there are no public results or independent tests included in this article. The company says that the agent only looks at the screen upon request, can limit itself to a single window, displays the steps it’s taking, allows you to stop a task, and must confirm destructive actions separately. These are the right safeguards, but so far they’re just stated capabilities. The price has not been finalized either; beta testers are promised an extended free trial period.
If such an agent were to appear on your computer tomorrow, what would you assign to it first: routine tasks in the browser or access to your email and files? Source: Hacker News - Newest: ""AI" "LLM""