A2UI-style in-chat UI runtime

Lets AI agents render interactive forms and cards inside conversations instead of just dumping text

Problem

Most AI is good at answering but poor at interacting with non-technical users; the common pattern is just 'AI asks, human answers', like the earliest BBS. Google's A2UI takes a low-code route that is a dead end for general cases, while Anthropic/OpenAI MCP Apps embed full web apps into chat, which is too heavy. Developers lack a lightweight middle layer that only renders interactive widgets.

Solution

An open-source runtime that lets agents declaratively render interactive components like forms, pickers, confirmation cards, and progress bars in the chat stream, then send structured results back. It deliberately avoids general low-code and focuses only on the narrow case where the agent needs user confirmation or choice.

Users

Frontend and full-stack developers building AI apps and agents, automation teams that need human-in-the-loop steps

MVP

Define 8-10 interactive components (form, choice, confirm, progress); provide JS/TS SDK and React components; serialize user actions to JSON back to the agent; adapter examples for mainstream agent frameworks; a runnable demo

Revenue

Open-source core plus hosted rendering subscription; enterprise edition with audit logs and permissions; marketplace revenue share

Why now

MCP and the agent ecosystem are exploding while the interaction layer is still fragmented ad-hoc code; an indie can enter with a small but beautiful declarative UI protocol