Mori

What it does

Mori is a pixel-cat desktop companion for Windows that helps people understand a screen area they deliberately select. Users can ask about that region and receive a plain-language explanation, a live note beside the selected spot, and a suggested next step while staying in control. Five Mozaik roles—Screen, Understanding, Action, Companion, and Memory—share request context; Memory does not save information automatically. Project presentation (PDF): https://drive.google.com/file/d/1E2-BLrXOLLDc8GZqzoJBweeTtM_XJSww/view?usp=drivesdk

How the agents run concurrently

Mori runs five concurrent roles inside a shared @mozaik-ai/core runtime. Each request publishes a shared event with a request ID, the user's prompt, recent chat, and selected-screen metadata. Screen records the chosen area; Understanding interprets it; Companion writes Mori's reply. Action and Memory can join the same request when enabled. The roles publish progress and results to shared request state and the live activity view. Screen, Understanding, and Companion can use selected image/video context; Action and Memory receive text only. Action only proposes a next step and never controls other apps. Memory can surface preferences explicitly stated in the current conversation but does not save them automatically.