openui

Why we included this project

OpenUI handles model-generated interfaces differently from the usual pattern: instead of asking the LLM for text or JSON, it works in a compact, streaming-first language built for structured UI, and renders the result live as tokens arrive. You bring your own component library, and OpenUI turns it into system-prompt instructions that define what the model may generate, so the components you already have are the ground truth for the output. The renderer then progressively paints charts, forms, tables, and layouts as the stream comes in. Because the language is compact, it claims up to 67% fewer tokens than JSON, which matters if you pay per token across many sessions. For small teams, the CLI scaffolding and ready-made chat interfaces get you to a working end-to-end setup quickly.