Skip to main content
cascadeflow integrates with the Vercel AI SDK as middleware, providing cascade routing for server-side AI applications with streaming support. The goal is still runtime intelligence inside the application loop: maintain streaming UX, keep tool execution server-side, and avoid a proxy hop on every model step.

​
Install

npm install @cascadeflow/vercel-ai

​
Quick Start

import { createChatHandler } from '@cascadeflow/vercel-ai'; import { CascadeAgent } from '@cascadeflow/core';  const agent = new CascadeAgent({  models: [  { name: 'gpt-4o-mini', provider: 'openai', cost: 0.000375 },  { name: 'gpt-4o', provider: 'openai', cost: 0.00625 },  ], });  const handler = createChatHandler(agent, {  protocol: 'data', // AI SDK v4 data stream  tools, // Tool definitions  toolHandlers, // Server-side tool execution  maxSteps: 5, // Multi-step tool loops });  // Use in Next.js API route, Express, or any Node.js server export const POST = handler;

​
Features

  • AI SDK v4 data stream and AI SDK v5/v6 UI streams
  • useChat multi-turn support — conversation history preserved
  • parts message format (AI SDK v6)
  • Tool call streaming visibility — see tool calls as they happen
  • Server-side tool execution via toolExecutor or toolHandlers
  • Multi-step controls: maxSteps, forceDirect
  • Cascade decision stream parts — optional metadata in the stream
  • Request-level overrides with allowlist + shared-secret guard

​
Why This Integration Matters

  • Preserves fast real-time UX while adding runtime control
  • Keeps governance logic close to tool loops and streaming responses
  • Avoids the compounded latency cost of external proxy mediation

​
Multi-Turn Chat

import { useChat } from 'ai/react';  export default function Chat() {  const { messages, input, handleSubmit, handleInputChange } = useChat({  api: '/api/chat',  });   return (  <div>  {messages.map((m) => (  <div key={m.id}>{m.content}</div>  ))}  <form onSubmit={handleSubmit}>  <input value={input} onChange={handleInputChange} />  </form>  </div>  ); }

​
Request Overrides

Override cascade behavior per request (protected by shared secret): 
const handler = createChatHandler(agent, {  protocol: 'data',  allowOverrides: ['forceDirect', 'maxSteps'],  overrideSecret: process.env.OVERRIDE_SECRET, });

​
Result

40-85% cost savings for Vercel AI SDK applications with streaming support and zero client-side changes.
Examples on GitHub: packages/core/examples/nodejs/vercel-edge.ts | vercel-ai-nextjs/