We are developing a new mobile operating system where agents are treated as a first-class primitive.
As an industry, we're rapidly developing next-generation hardware to host the new software paradigms created by LLMs and other foundational models. Current hardware relies on cloud-based AI infrastructure or solutions siloed in the application-layer, lacking deep integration with the user. Instead of creating new hardware, we’re building agentic infrastructure into the devices you already use.
We have two core theses on what incorporating agents into the lowest-level implementation layer unlocks:
-
Predictive > Prompted
When designing agentic systems, access to data is the most important factor in creating a compelling user experience. With enough data, we can predict and autocomplete actions for the user rather than rely on them to supply essays of prompting. Implementing at the OS layer allows us to build valuable user context across app intersections by reading the databases that are typically sandboxed within apps and seeing user interaction patterns across these app sessions. All of this context allows an agent to make a more informed decision on when and how to perform an action.
By giving an agent the same hooks into a mobile device that any user would have, we unlock a seemingly infinite action space for a model to operate within. The interfaces users manipulate to operate our devices are the same ones our agents use; these include a virtual keyboard, touchscreen, volume buttons, and the power button. We include additional efficiency hacks afforded by OS-level code to avoid common pitfalls when navigating UI hierarchies.
OS-level agents will never require manual integrations with third-party apps, and they're not limited by platforms not exposing a public API. This means that agents can drop into a rich environment of apps that people are already using and become effective immediately.
We believe that by combining these central themes, we can make the best piece of AI-native hardware the device that's already in your hands.