Features

Everything you need to go hands-free

GestureOS pairs a state-of-the-art vision pipeline with a context engine that knows what you're doing — and what you mean.

A palm detector finds your hand, then a landmark net maps 21 3D joints every frame — the same skeleton you see in the live overlay.

ONNX models execute locally with DirectML / CUDA acceleration. No frames ever leave your machine. No latency, no cloud, no creep.

A Kalman-filtered tracker plus async inference keeps gestures buttery at 60fps, even while the model crunches in the background.

GestureOS detects the focused app and swaps gesture maps automatically — gaming, meetings, and system control each get their own bindings.

Temporal voting, hysteresis, and ghost-gesture suppression mean it only fires when you mean it. Wave goodbye to phantom clicks.

Watch what the engine sees in real time. Every joint, every confidence score, rendered as a glowing skeleton you can trust.

Gesture vocabulary

Context profiles

Gaming

Map gestures to WASD, abilities, and camera control. Lean into a game without ever touching the keyboard.

Meetings

Mute, raise hand, and toggle camera with a flick. React with emoji gestures your teammates actually see.

System control

Scroll, switch windows, adjust volume, and navigate your desktop hands-free — perfect for presentations and accessibility.

Under the hood

GestureOS isn't a demo — it's a tuned production pipeline.

Landmark positions are smoothed by a Kalman filter and velocity-clamped interpolation, so gestures stay steady even when frames are noisy.

Binary gates were replaced with weighted confidence scoring, letting precision and recall be tuned per gesture independently.

Temporal voting, hysteresis, and an intent latch ensure an action only fires when you truly mean it — no phantom clicks.

Heavy ONNX inference runs on a dedicated thread feeding a ring buffer, keeping capture and overlay smooth at 60fps.