The Poincaré Manifesto

Computing has always advanced by reframing the relationship between people and machines. Each generation invented a new interface that changed what felt natural to ask from a computer. But for all the progress in processors and neural engines, today’s smartphones still behave like artifacts of an earlier age. They wait for commands, forget what happened a moment ago, and fragment every task into isolated apps that compete for attention instead of cooperating to solve problems.

We believe this gap is no longer acceptable. Intelligence is now native to the device, yet the operating system treats it as an accessory. Users still orchestrate their own workflows by tapping, navigating, and re-entering context. The machine sits idle, even though it is capable of interpreting, predicting, and executing.

OS-A is our response. It is a full Agentic operating system designed for a world where intelligence is not an add-on but the organizing principle. Instead of being an app launcher, the OS becomes a system that understands intention, maintains context, and coordinates action across the entire device.

At the center of the OS is a model of the user’s world that lives entirely on the device. This model is not a search index or a history log. It is a continuously evolving representation of routines, commitments, documents, relationships, and patterns of behavior. It lets the OS recognize when two tasks are related, when an action is part of a larger plan, and when a request can be completed without interrupting the user. The result is a phone that can support long arcs of activity rather than isolated steps.

To execute those tasks, OS-A uses a network of tightly constrained micro-runtimes that can read interfaces, call tools, interact with apps, and complete workflows end to end. These runtimes behave like cooperating processes that share a unified understanding of user context. They adapt to real conditions on the device, resolve ambiguity on the fly, and adjust trajectories when tasks evolve. This architecture allows the system to operate across the existing ecosystem while gradually reducing the need for explicit app boundaries.

Our research explores how such systems can become more stable, more predictable, and more trustworthy over time. We work on representations that allow the OS to compress long-term user behavior into meaningful state. We study how multiagent coordination can remain controllable while still flexible. We investigate ways to design interfaces that recede when not needed and reveal structure when guidance is valuable. And we focus on keeping all of this intelligence private, local, and under the user’s control.

Poincaré Labs is an applied research and product group committed to inventing the foundations of agentic computing. Our work lies at the intersection of operating systems, embodied AI, long-term context, and interface design. We are building systems that treat intelligence as a substrate, not a feature stitched onto legacy frameworks.

The future of computing will belong to devices that understand goals, preserve context, and act with autonomy. It will belong to operating systems that take responsibility for execution so people can take responsibility for thinking.

OS-A is our step toward that future.