Universal AI streaming infrastructure

One wire protocol. Any server. Any client framework.

The RAIS Protocol (React AI Stream) is a minimal three-event SSE standard for streaming AI responses from any backend to any frontend. This monorepo ships everything you need: a React hook, Vue composable, Express middleware, Python helper, DevTools panel, and a scaffolding CLI.

npm install @react-ai-stream/react

Try it live — right here

No API key. No install. Streaming starts instantly.

Full playground with protocol inspector →

30 seconds to a streaming chat

npx create-ai-stream-app

Or install manually:

'use client'
import { useAIChat } from '@react-ai-stream/react'
import { Chat } from '@react-ai-stream/ui'
import '@react-ai-stream/ui/styles'
 
export default function Page() {
  const { messages, sendMessage, loading, stop } = useAIChat({
    endpoint: '/api/chat',   // OpenAI, Anthropic, Groq, FastAPI — any streaming endpoint
  })
  return <Chat messages={messages} onSend={sendMessage} onStop={stop} loading={loading} />
}

The hook returns plain data. Drop <Chat /> in for zero-config, or wire messages to any UI. Full quickstart →

All packages — click to explore

Who is this for?

Why backend-agnostic matters

Most AI chat libraries are secretly backend libraries. They stream from OpenAI directly, or through their own cloud, or via a specific server adapter. The React hook is just a thin client on top of one particular provider.

react-ai-stream takes a different approach: the hook speaks a simple HTTP streaming protocol. Any server that produces that protocol works.

data: {"type":"text","text":"Hello"}
data: {"type":"text","text":", world"}
data: {"type":"done"}

Three event types. That's the entire contract between your server and your React component. This means:

• Switch providers without touching React. OpenAI → Anthropic → your own model? Change the API route, not the frontend.
• No API keys in the browser. The hook talks to your server, which talks to the LLM.
• Multiple providers simultaneously. Run useAIChat three times with three endpoints — each instance is fully isolated.
• Any backend language. FastAPI, Go, Rails, Cloudflare Workers — if it can stream SSE, it works.

Deep dive: Why backend-agnostic? →

What you get

The RAIS wire protocol

Any server in any language that emits these three event types works with any RAIS client:

Content-Type: text/event-stream

data: {"type":"text","text":"Hello"}

data: {"type":"text","text":" world"}

data: {"type":"done"}

That's the entire protocol. Verify any server with:

npx rais-compliance http://localhost:3000/api/chat

Full protocol spec → · Compliance checker →

Next steps