Read the visible world
Interpret changing screen content and interface states as a visual environment.
AI-native screen perception
Give AI a clearer view of the screen.
Syrup is building a perception engine that helps AI systems understand visual interfaces as structured, actionable environments.
Starting with a working MapleStory proof of concept.
⌘
Screen → structureVisual context, made usable01 pixels
02 interface signals
03 context
04 actions
The problem
AI can act on a screen only as well as it can read one.
Interfaces are dense, dynamic and often lack clean machine-readable structure. Syrup is focused on the perception layer between what appears on screen and what an AI system can reason about and do.
Turn pixels into context
Organize visual signals into representations that are useful for downstream reasoning.
Support grounded action
Give AI systems a clearer basis for interacting with software through its interface.
First proof of concept
MapleSyrup puts the idea on screen.
The first working exploration applies Syrup’s screen-perception approach to MapleStory. It is an early proof of concept for investigating how an AI system can interpret a live visual environment.
Working POC · Early stage
MapleSyrup
A focused prototype exploring screen understanding in a visually rich game environment. The work is ongoing; the prototype is a technical starting point, not a production product.
View the public code ↗Download the Windows installer ↓Download the portable ZIP ↓MapleSyrup 0.8.0 · Windows 10/11 64-bit · installer and portable build
Our approach
Build the perception layer with engineering discipline.
We are starting with a concrete environment and a systems-level mindset, then building toward a reusable foundation for screen-based AI interaction.
Start with observable behavior
Ground development in a real interface and testable visual signals.
Keep the system modular
Separate perception from the application-specific logic built on top of it.
Measure before making claims
We are early and will share validated milestones as they are established.
Design for practical use
Focus on useful context that can support reliable downstream decisions.
Get in touch
Interested in screen perception?
We’d like to hear from builders, researchers and early partners exploring how AI can understand and operate visual software.