Hands full. Still in control.

This is an experiment on the layer: Praxis defines multimodal human interaction as one coherent product direction, and retires the assumption that the only accepted posture is sitting at a desk operating a web app.

What it is

Praxis models interaction as practiced action:

human signal + context + permission + feedback = governed action

The signal may be a spoken phrase, a controller chord, a wrist motion, a hand gesture, a camera-observed posture, a gaze target, or a visual selection. The result is the same kind of thing Unifyhub already understands: an intentional, authorized action with an audit trail.

Per the durability principle, Praxis is defined by device classes rather than any vendor's hardware: watches, wristbands, depth cameras, controllers, and microphones are interchangeable sources for the same governed flow.

The interaction flow

signals (voice, buttons, wearables, vision)
  → signal buffer (time + source + confidence)
  → aggregation (one candidate intent)
  → context (user, place, target, mode)
  → intent frame
  → identity + policy
  → action mapping
  → dispatch
  → feedback

Signals

A signal is any raw input Praxis can interpret: spoken words, button presses, joystick movement, wrist motion, camera observations, hand gestures, gaze, proximity, or UI selection. Signals become actions only through the flow above. A controller button may mean "select" in one mode, "confirm" in another, and nothing at all in a restricted context.

Aggregation

Aggregation turns several partial signals into one candidate intent, and the same action may be assembled from several sources at once:

• a watch on one wrist reports motion, orientation, and a confirmation tap

• a gesture wristband on the other reports pressure or low-motion hand intent

• a depth camera observes the scene, target position, hand location, or posture

• voice or controller input names the action, target, mode, or parameter

Praxis centralizes these into a time-aligned signal buffer, compares their confidence, discards noise, and emits one intent-frame candidate. The receiving product never needs to understand every sensor; it receives one governed request with clear provenance.

Intent frame

The intent frame is the structured object Praxis builds before asking Unifyhub to do anything. It carries: who is acting, which modalities contributed, which device streams contributed and when, what target the person appears to mean, what action is requested, how confident the system is, what confirmation level is required, and what audit context is retained.

Mapping

Mappings translate intent frames into product-specific actions. A mapping can be simple ("controller X opens the active panel") or rich ("a voice command plus a wrist rotation changes the selected device mode"). Mappings are explicit and inspectable, like every other policy in the system.

Confirmation

Confirmation is proportional to risk:

• Low-risk, reversible: happens immediately, with feedback.

• Medium-risk: asks for lightweight confirmation.

• High-risk or irreversible: requires explicit, policy-backed confirmation and may require another role or surface.

Feedback

Every action returns feedback in the modality the person can use at that moment: voice response, haptic tap, controller vibration, visual state, audit entry, or a combination. Good feedback answers three questions: what did Praxis understand, what did Unifyhub do, and what can the person do next.

Who it is for

• Operators who need to move, inspect, acknowledge, steer, or act away from a desk.

• Builders who want to prototype new interaction devices and map them to real workflows.

• Supervisors who need clear confirmation, audit, and escalation when embodied control touches risky actions.

• People with different access needs, for whom voice, buttons, motion, and visual feedback can substitute for each other.

All products · Roadmap · All principles