AgentOS: the operating system that thinks for you

We didn't fork Android. We didn't build on Wear OS. We wrote an operating system from scratch, designed specifically for agentic AI on a wearable form factor.

When we started Heymacy, the easy path would have been to build on Wear OS or fork AOSP like every other smartwatch company. We considered it seriously. Then we realized that an operating system designed for phone-style app grids would never work for an agentic device. So we built AgentOS from the ground up.

Why existing watch OS platforms don't work

Wear OS, watchOS, and Tizen were all designed with the same assumption: the watch is a small phone. They use app launchers, notification trays, and tiny touch targets. This makes sense if your watch is a notification mirror, but it completely falls apart when the watch needs to be autonomous and intelligent.

An agentic OS needs to be fundamentally different. It needs to run AI models locally with low latency. It needs to maintain persistent context about the user across sessions. It needs to manage multiple concurrent tasks (listening, sensing, acting) without draining the battery in two hours. None of the existing platforms were designed for this.

The architecture of AgentOS

AgentOS is built on three layers. At the bottom is a lightweight real-time kernel optimized for our hardware. In the middle is the Agent Runtime, which manages the AI models, sensor fusion, context graph, and task execution. On top is the Experience Layer: the expressive watch face, voice interface, and haptic feedback system.

The Agent Runtime is the heart of AgentOS. It maintains a context graph that combines everything Macy knows: your calendar, location, health data, habits, preferences, and the current environment (noise level, motion, time of day). When you make a request or when Macy proactively suggests something, it draws on this full context to make intelligent decisions.

On-device AI, not cloud-dependent

We run quantized language models directly on the watch. With 8GB of RAM and a dedicated neural processing unit, Macy can understand natural language, generate responses, and make decisions without sending your data to a cloud server. Internet connectivity enhances capabilities (real-time information, service integrations) but isn't required for core intelligence.

The expressive face

One of the most distinctive features of AgentOS is the animated face. Macy has eight emotional states that it displays through expressive eye animations on the AMOLED screen. It blinks when idle, narrows its eyes when thinking, lights up when it has a suggestion. This isn't a gimmick. Research shows that people interact more naturally with devices that have personality. The face turns Macy from a tool into a companion.

What's next for AgentOS

We're building AgentOS as an open platform. Third-party developers will be able to create "skills" that plug into the Agent Runtime, giving Macy new capabilities without the overhead of traditional apps. Imagine a skill that monitors your glucose levels and proactively adjusts your meal suggestions, or one that manages your crypto portfolio based on rules you've set. The platform is designed to grow with the community.