HazelJS vs Vercel AI SDK

TL;DR

• Vercel AI SDK shines for streaming UI, chat hooks, and Vercel-centric route handlers.
• HazelJS is a backend-first TypeScript framework: controllers, DI, agents, RAG, memory, and microservices.
• Use the AI SDK for frontend streaming UX (often with any backend).
• Use HazelJS when the hard problem is the server: agents, tools, RAG, auth, and durable workflows.

Who this is for

Choose this comparison if you:

• Build Next.js apps and wonder whether AI SDK alone is enough for production agents
• Need multi-provider LLMs, RAG, and HITL on the server
• Want deployment beyond Vercel (Lambda, containers, Cloud Functions)

Prefer Vercel AI SDK alone when:

• The product is mostly chat UI streaming against simple route handlers
• You are all-in on Vercel and do not need a Nest-style backend framework

Side-by-side

Complementary, not always exclusive

Many teams use:

• Vercel AI SDK on the client for useChat / streaming UX
• HazelJS (or Nest) on the server for agents, RAG, and business tools

HazelJS competes when you expected the AI SDK backend pieces to replace a full agent/RAG platform — they usually do not.

Code: route handler vs framework agent

Typical AI SDK route (illustrative): stream text from a model in a serverless handler — great for chat UI, limited as an agent OS.

HazelJS agent + HTTP:

import { Agent, Tool } from '@hazeljs/agent';
import { Controller, Post, Body } from '@hazeljs/core';

@Agent({
  name: 'desk',
  enableMemory: true,
  enableRAG: true,
  systemPrompt: 'Help users with account questions.',
})
export class DeskAgent {
  @Tool({
    description: 'Fetch account plan',
    parameters: [{ name: 'userId', type: 'string', required: true }],
  })
  async plan(input: { userId: string }) {
    return { userId: input.userId, plan: 'pro' };
  }
}

@Controller({ path: '/desk' })
export class DeskController {
  @Post('message')
  async message(@Body() body: { text: string }) {
    // Execute via Agent Runtime — see agent package docs
    return { received: body.text };
  }
}

Decision guide

When Vercel AI SDK is better

• Shipping UI fast on Vercel is the bottleneck
• You do not need durable agents, GraphRAG, or microservices patterns
• Team skill is React/Next, not backend frameworks

Related

• HazelJS vs LangChain
• HazelJS vs NestJS
• Agent Runtime
• Serverless guide

Next steps

1. Installation
2. AI package
3. Support agent example
4. GitHub