Sayhi-bzb/Agent-HTML

You don't need a chat ui but a canvas with ai.

What it solves

Agent-HTML addresses the limitations of Markdown for AI agent outputs. While Markdown is suitable for text and code, it cannot handle complex visual needs like dashboards, interactive data views, product mockups, or themed reports. This project allows agents to generate durable, interactive React artifacts instead of simple text documents.

How it works

The system uses a "Canvas" architecture consisting of several parts:

  • agent-html: A portable workspace where agents write React and TypeScript artifacts.
  • @agent-html/react: A protocol providing headless Artifact and Block components that create stable, addressable regions within the HTML surface.
  • Canvas Host: A Vite-powered environment that renders artifacts, provides validation diagnostics, manages theme presets, and handles block-level prompt routing.
  • Local Resources: A set of UI primitives, hooks, and schemas that live alongside artifacts to ensure consistency.

Who it’s for

It is designed for developers building AI agent workflows where the output needs to be a functional user interface—such as operational dashboards, Kanban boards, or data reports—rather than just a chat response.

Highlights

  • Inspectable Surfaces: Uses Block metadata to allow targeted review and focused feedback on specific regions of an artifact.
  • Rich Expression: Supports layouts, state, interaction, charts, and tables.
  • Durable Context: Stores rules, resources, and artifacts in the filesystem to prevent them from being lost in chat history.
  • Targeted Revision: Enables follow-up prompts to be routed to specific blocks of the HTML surface for precise editing.

Related

  • Project
    stackblitz/bolt-slides

    A framework for creating presentation decks as interactive web apps, enabling AI agents to generate live, responsive slides with 3D and real-time data from a single prompt.

  • Project
    plannotator/effective-html

    A collection of agent skills that enable AI agents to generate self-contained HTML artifacts like wireframes, prototypes, and diagrams for better visual communication.

  • Dispatch
    The Unreasonable Effectiveness of HTML in the Age of AI Agents

    Exploring why developers are shifting from Markdown to HTML for AI-generated artifacts, balancing interactive power against token costs and maintainability.

  • Project
    scaleapi/scale-agentex

    A framework for building and deploying AI agents that scales from simple chatbots to fully autonomous systems, supporting both synchronous and asynchronous workflows.

  • Project
    get-convex/agent

    A framework for building AI agents on Convex with persistent chat threads, tool calling, and real-time streaming updates.