Every installed application is an orchestrator of OS capabilities. LLMs are simply a more expressive interface for that orchestration. Zuzu is a framework for building installable, AI-native desktop apps where the intelligence runs on the user's hardware — not in a data center. It uses JRuby and Glimmer DSL for SWT for the GUI, Mozilla's llamafile for local LLM inference, and SQLite (via AgentFS) as a sandboxed virtual filesystem the agent can read and write without touching the host OS. Apps ship as a cross-platform .jar, or as a native installer (.dmg/.deb/.exe) with a JRE bundled via jpackage — users download, double-click, and run with no Java installation required. No cloud. No subscriptions. No infrastructure to operate. Scaffolded projects include CLAUDE.md and Claude Code skills pre-tuned to Zuzu's patterns, so coding agents generate correct framework code from the start.