Why S1 is not a smartwatch
Layering AI apps onto a smartwatch and redesigning a terminal around AI Agent are two very different projects.
Generative AI is moving from software tooling towards personal agents and device-cloud terminals. Phones, smartwatches and earbuds all have some voice capability, but their interaction paths, power strategies, sensor choices and product logic were designed around conventional app ecosystems. None of them is properly adapted to what an AI Agent actually needs: persistent availability, proactive service, multimodal input and long-term memory.
So the starting point for S1 was never "build a smarter watch."
The problem we set out to solve
People need an AI entry point that is lighter and more natural than operating a phone, while being more mobile and more private than a smart speaker. The phone's problem is the weight of the interaction path: take it out, unlock it, find the app, open it — only then can a conversation begin. The speaker's problem is that it does not travel with you, and offers no privacy in a shared room.
A wearable terminal can handle voice input, image capture, status feedback and urgent interruption without occupying your hands. That is exactly what personal assistance, on-the-go capture, mobile work and field support scenarios actually require.
Three concrete trade-offs
No complex apps, no frequent touch. S1 uses a low-power, non-touch e-paper display for battery level, network status, confirmations and short summaries. It is not a screen for scrolling feeds; it is a low-interruption window that tells you the system understood you.
Keep physical interruption. A flush-mounted AI Button on the top edge starts a conversation and interrupts the current reply. Relying purely on voice interruption is unreliable in real noise, and a physical key you can always press is more certain than any amount of wake-word tuning.
Power the camera on demand. The front camera exists for photographing and recognising objects, not for a continuous video stream. That is a power decision and a privacy decision at once: data you never capture cannot leak.
Where the project stands
Product positioning, overall architecture, structural definition and core component selection are complete, formal schematic design is under way, and the project has entered the DVT-ready EVT stage. Future posts will unpack the AI Core + Dock modular architecture, and the hardware tolerance work aimed at real-world speech.