The web UI for AI agents
Framework-agnostic web components
Build AI-native apps that run anywhere.
The web UI layer for AI agents. Bring your model, SDK, or your own agent loop — drop in framework-agnostic components for streaming, tool calls, reasoning, and artifacts, in any app.
npm i @kitn.ai/ui One component model — every framework
- React
- Vue
- Svelte
- Angular
- Solid
- HTML
You bring the model.
AI/UI brings the interface.
Streaming, tool calls, reasoning, generative-UI cards, artifacts, and voice — the agent UI, already built and battle-tested. Wire up your model or harness and ship.
Built for
AI-native apps
AI/UI is the web UI layer for AI agents and harnesses — streaming, tool calls, reasoning, generative-UI cards, and artifacts. The hard parts, already built as framework-agnostic web components you wire to your model or agent loop and drop into any app.
Browse componentsConversation
Token-by-token streaming, rich markdown, code highlighting, and message actions. The chat surface, done right.
Agentic UI
Tool calls and reasoning render inline — show the work as your agent thinks, calls tools, and acts.
Generative UI
Confirmations, choices, tasks, and forms render as typed, interactive cards the model can drive.
Artifacts & files
Preview live code, docs, and sites in a sandboxed panel — plus drag-and-drop files and image input.
Your harness has a backend.
This is the front-end.
Most agent harnesses are CLI or headless — the loop, the tools, and the model wiring are already done, but there's no web UI. AI/UI is that layer: framework-agnostic web components you point at your harness's stream. Bring your own loop; we render the chat, tool calls, reasoning, and artifacts.
Wire it to your agent- PiOpen-source agent harness + coding CLI
- OpenClawHeadless coding agent (ACP)
- Bring your own. Any agent loop, SDK, or model — AI/UI just needs a stream to render.
Ship any AI surface — not just a chatbox.
Build a real AI workspace
A collapsible conversation sidebar, resizable panes, and a file tree — the shell for a coding agent or an internal copilot.
See the example
Answers that show their receipts
Stream grounded responses with sources and citations, so users can trust — and verify — what the model says.
See the example
Show the work
Collapsible reasoning traces stream alongside the answer — perfect for debugging copilots and analyst agents.
See the example
Render what the model makes
Preview generated sites, code, and documents in a sandboxed, resizable panel — the Claude- and ChatGPT-style canvas.
See the exampleUp and running — no build step.
The fastest path is a single <script> tag from a CDN — paste it
in and the custom elements register themselves. Prefer a bundler? Install from npm.
Either way, set messages in JavaScript to wire up your model.
- 01
Add the script
<script type="module" src="https://cdn.jsdelivr.net/npm/@kitn.ai/ui/dist/kitn-chat.es.js" ></script> - 02
Drop one in
<kai-chat></kai-chat>
- 01
Install
npm i @kitn.ai/ui - 02
Register the elements
import '@kitn.ai/ui/elements'; - 03
Drop one in
<kai-chat></kai-chat>
Match your brand — without forking anything.
Every color, font, radius, and shadow is a CSS custom property. Override a
handful of --kai-* tokens and the whole UI
reskins, light and dark. Or design it live: 40+ presets, font and shape
controls, and export-ready CSS you can paste straight into your app.
Start building your AI interface.
Open source, MIT-licensed, and ready for production. Install it, drop in a component, and wire up your model.
npm i @kitn.ai/ui