AI Technology
One Architecture for Movement and Intelligence
COMMERCELINK links real-world motion sensor feeds to AI models that interpret what a player does and respond within the game. This page explains how that stack fits together and which parts are still in development.
Model Layer
Two Model Roles, One Game Loop
Multimodal understanding and dialogue are handled by different model families, each assigned to the part of play it serves best.
Google Gemini
Gemini is used for multimodal understanding, interpreting combinations of movement, images, and contextual inputs so the game can reason about its physical and spatial surroundings.
Claude Models
Claude is used for dialogue, storytelling, and contextual responses, supporting character conversation and branching narrative that adapts to what the player has done.
Inference Stack
From Sensor Signal to Game Response
The core pipeline is built from four layers. Each layer is designed to handle a distinct part of the interaction.
Motion Recognition
Pipelines interpret movement, gestures, and motion-sensor data as gameplay input.
Adaptive Gameplay
Missions and scenarios adjust in response to how the player moves, speaks, and interacts.
AI Inference & Cloud
Edge inference handles latency-sensitive input, while cloud infrastructure supports heavier model workloads.
Development Status
Where Each Layer Stands
Features below are in active development unless confirmed otherwise. Nothing here describes a released capability.
01
Motion recognition pipeline: in development
02
Gemini and Claude integration layer: in development
03
Cloud inference and adaptive loops: in development