Agentastic: one personal AI agent on iMessage, WhatsApp, mobile, web, and every app on your Mac — one memory that remembers everything, picking up wherever you left off.

You are halfway through an email. “Make this warmer” refers to the sentence under your cursor, the person receiving it, the meeting you had last Tuesday, and the fact that you are trying not to sound annoyed. None of that fits neatly into a prompt.

So you copy the paragraph. Open another app. Reconstruct the situation. Ask for a rewrite. Carry the answer back. Then do it again for the calendar invitation, the research question, the presentation, and the follow-up you almost forgot to send. The model is intelligent. The workflow is not.

We believe the human-agent interface is one of the primary bottlenecks in making agents universally accessible and useful. Agentastic begins with a simple premise: the agent should come to you. Call it from any application. Type a request inline: [make this warmer]. Highlight the text, image, or page you mean. Open the full interface when the job needs more room. Text it on iMessage, WhatsApp, pick it up on mobile or the web, or reach it in Slack or Telegram — wherever the thought occurs. However you reach it, you are not starting over.

The agent can gather and understand the immediate context (the application, the screen, the browser, the document) as well as the larger context behind it: the project, the people, the previous decisions, and the way you prefer the work to be done.

But familiarity without agency is only a more pleasant chatbot. Agentastic is built to do the work. Ask for research and it can browse, investigate, compare sources, and write the report. Ask for a presentation and it can build the file. Ask for a site and it can design and implement it. It can operate software, create documents and spreadsheets, generate images, and use the computer when the work cannot be completed through an API.

Chat is a fine interface for a conversation and a poor one for most work. Agentastic is built on generative interfaces: it presents each piece of work in the shape it calls for, whether that is a report, a form, a dashboard, a site, or a small application that did not exist a minute ago.

Underneath, Agentastic is not one enormous brain pretending to be good at everything. It is a collection of models that decide when to get involved and what to contribute. Different models are good at different kinds of work. Different agents need different tools, instructions, personalities, and areas of responsibility. Our goal is to build a collective intelligence where agents work side by side with humans, learning to coordinate and communicate with each other as well as with their human partners.

Agentastic is our first experiment in learning what the future of human-agent collaboration looks like: how to design agents that understand us better, have better context about the work they do, and know when to interrupt and when to be quiet. How to make agents better partners.