omdsh-dev/dsh-genui

What it solves

dsh-genui transforms static text replies from AI models into interactive, visual user interfaces. Instead of receiving a plain text answer or a raw code block, users see rendered components—such as dashboards, charts, and interactive forms—embedded directly within the conversation flow, making complex data and interactions more intuitive and actionable.

How it works

The project uses a declarative JSON specification. The AI model writes a description of the UI (e.g., a table or a chart) inside a special dsh-ui code fence. A browser-side renderer then intercepts this JSON and converts it into a safe, whitelisted set of UI components. To maintain performance, the core renderer is lightweight, while heavy engines like ECharts, Mermaid, and Three.js are lazy-loaded only when needed.

Who it’s for

It is designed for users and developers using the DSH ecosystem who want to move beyond text-only interactions with LLMs and create rich, application-like experiences within a chat interface.

Highlights

• Answer-as-UI: Components render in real-time as the model streams its response.
• Extensive Component Library: Over 30 components including stat cards, sortable tables, ECharts, Mermaid diagrams, and 3D-oriented elements.
• Local-First Interaction: UI state changes (like quiz grading, tab switching, and parameter sliders for plots) happen instantly on the client side without needing a model round-trip.
• Persistent Session Panels: Supports a resizable dock for persistent work surfaces that can be updated in place.
• Safe Rendering: Uses a strict whitelist and a standalone parser for function expressions to prevent the execution of arbitrary scripts.