Figma Plugin API 2.0 adds on-device model hooks for safer prompt processing

Design · 3 min read

Figma rolled out Plugin API 2.0, which introduces on-device model hooks that allow plugin authors to run limited-size ML models locally inside the Figma desktop app. This is designed for operations like prompt pre-processing, simple style inference, or token anonymization without leaving the user’s machine.

The new API also offers a standard pattern for hybrid plugins: lightweight on-device steps followed by optional server-side processing for heavier compute. Fallback logic and clear permission prompts give users control over when assets leave their devices.

Figma published migration guides and supply examples like a plugin that infers component roles from frames and annotates them locally before sending only metadata upstream. The move aligns with growing demand for privacy-conscious AI features in design tooling.