Vercel AI SDK · Course Guide

This course spans 22 chapters (including 3 capstones), based on the latest official documentation at ai-sdk.dev: chapter structure, APIs, and example code align with the official docs; explanatory text translated to English.

What You Will Learn

Stage	Chapters	Content
Foundation	01–04	SDK architecture overview, quick start, provider management, prompts and message models
Core	05–15	Text generation & streaming, structured output, tool calling & MCP, reasoning models, embeddings & reranking, middleware, telemetry, multimodal & speech
UI	16–17	useChat chat interface, UI Message Stream, message persistence
Agent	18–19	ToolLoopAgent, loop control, approval, memory, and subagents
Capstone	20–22	RAG knowledge-base Q&A → multi-agent customer ticket system → full-stack streaming chat app

Provider Compatibility Note ⭐

All example code in this course is unifiedly compatible with both access methods — switching models requires changing only one place:

// Method 1: Vercel AI Gateway (recommended default)
import { createGateway } from 'ai';

const gateway = createGateway({
  apiKey: process.env.AI_GATEWAY_API_KEY ?? '',
});

const model = gateway('openai/gpt-5'); // gateway/model-id format

// Method 2: Custom OpenAI-compatible Provider
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';

const myProvider = createOpenAICompatible({
  name: 'my-provider',
  baseURL: process.env.OPENAI_COMPATIBLE_BASE_URL ?? '',
  apiKey: process.env.OPENAI_COMPATIBLE_API_KEY ?? '',
});

const model = myProvider('gpt-4o-mini'); // model name as your server actually provides

💡 Throughout the course, we never directly depend on vendor-specific packages like @ai-sdk/openai or @ai-sdk/anthropic — this is the essence of the AI SDK's "provider-agnostic" design. Chapter 3 covers the provider registry, aliases, and routing in depth.

Official Reference Sources

Source	Link
Official docs	ai-sdk.dev
GitHub repo	vercel/ai
Cookbook	ai-sdk.dev/cookbook

Content version baseline: AI SDK v7 (npm ai@7.0.x, cross-referenced with the latest ai-sdk.dev docs).

Learning Recommendations

• Chapter 3 (Provider Management) is the foundation of the entire course: all models in subsequent chapters are constructed through it;
• Chapters 5–8 are the core daily-development APIs: generateText / streamText / generateObject / tool;
• The Agent duology (Chapters 18–19) progresses step by step: learn to run the loop first, then control approval and memory;
• The three capstone chapters can be read independently: each is a complete, runnable project.

👉 Start with Chapter 1.