Web chat built on eve and TanStack Start.
Build an AI agent that presents answers as interactive components in a web chat. This TanStack Start template connects an eve agent to a React frontend, with json-render turning structured responses into result cards, sortable tables, metrics, and step-by-step layouts.
Use it as a starting point for a research assistant or a conversational interface for your own tools. The included agent can search the web, read pages, and ask follow-up questions when it needs more information.
Generative UI lets the agent choose how to present an answer using a catalog of components you define. With json-render, the agent produces a structured specification that your React components render.
The template validates component properties before displaying the result. You can extend the catalog with your own components and keep the presentation consistent with your app. For example, a product comparison can become a sortable table, and a travel plan can become an ordered itinerary.
The chat streams responses and displays tool activity alongside messages. Each conversation has a URL that lets you return to the session after reloading the page.
Question forms let people choose an option or type an answer without leaving the conversation. An included approval demo shows how a tool can wait for a person's decision before continuing.
Edit the agent's instructions to define its role, add tools for your services, and customize the chat's React components. The frontend uses shadcn/ui and Tailwind CSS, with TanStack Table providing sortable data tables.
Browser access uses a shared password. You can replace that flow with your own authentication provider as you develop the app.
The TanStack Start frontend and eve agent deploy together in one Vercel project through Vercel Services. Deployed agent sessions use Vercel Sandbox.
Set the EVE_ACCESS_PASSWORD environment variable during setup, then open the deployed app and sign in to start a conversation.
