SYS/07 — AI INFRASTRUCTURE
ZXERO
AI operating layer — memory, perception, and action across the whole machine.
- STATUS
- In active development
ZXERO sits above every app on the machine. It's built on three pillars: memory that persists across tools, perception of what's on screen and active, and the ability to execute rather than only answer.
KEY FACTS
- Three pillars: persistent memory across tools, screen perception via accessibility APIs and screenshots, and the ability to execute rather than only answer.
- Model-agnostic — routes tasks across OpenAI, Anthropic, Gemini, Perplexity, and local models rather than being tied to one provider.
- Interface is a native macOS desktop companion — push-to-talk, voice, global shortcuts, screen understanding.
- Status: in active development.
The problem
Context loss. Every AI session starts from zero. Users spend their time copying, pasting, and re-explaining the same work between ChatGPT, Claude, Cursor, the browser, and everything else.
How it works
Three pillars. Memory: persistent context that follows the user across ChatGPT, Claude, Gemini, Cursor, VS Code, the browser, docs, and projects. Perception: reads the screen, detects active applications and windows, understands UI via accessibility APIs and screenshots. Action: executes rather than only answering -- opening apps, filling forms, editing documents, writing code, running shell commands.
Model-agnostic, and where it runs
Routes tasks across OpenAI, Anthropic, Gemini, Perplexity, and local models rather than being tied to one provider. The interface is a native macOS desktop companion -- push-to-talk, voice, global shortcuts, screen understanding. Positioned as AI as operator, not chatbot.
Vision
This section is explicitly aspiration, not current capability. The long-term ambition is a general-purpose "digital operator" -- orchestrating full workflows (research, build, test, deploy) and eventually enterprise use cases. None of that exists today; it's the direction, not the product.