Vercel AI SDK & Full-Stack Patterns
From the course TypeScript for AI Applications
The Vercel AI SDK
Core Concepts
The AI SDK provides a unified interface for AI providers with built-in streaming:
import { generateText, streamText } from "ai";
import { openai } from "@ai-sdk/openai";
// Non-streaming
const { text } = await generateText({
model: openai("gpt-4o"),
prompt: "Explain React hooks",
});
// Streaming
const result = streamText({
model: openai("gpt-4o"),
prompt: "Explain React hooks",
});
Structured Output with AI SDK
import { generateObject } from "ai";
const { object } = await generateObject({
model: openai("gpt-4o"),
schema: z.object({
recipe: z.object({
name: z.string(),
ingredients: z.array(z.string()),
steps: z.array(z.string()),
}),
}),
prompt: "Create a recipe for chocolate chip cookies",
});
// object is fully typed as { recipe: { name: string; ... } }
Provider Switching
import { anthropic } from "@ai-sdk/anthropic";
import { google } from "@ai-sdk/google";
// Same interface, different providers
const result = await generateText({
model: anthropic("claude-sonnet-4-20250514"), // or google("gemini-2.5-flash")
prompt: "Hello",
});
Full-Stack Pattern
- Server: AI SDK for provider abstraction and streaming
- Shared: Zod schemas for input/output validation
- Client: React hooks for streaming UI
- Testing: Mock providers for deterministic tests
Common Pitfall: Importing Provider SDKs Directly in Shared Code
Importing openai or @anthropic-ai/sdk directly in shared utility functions couples that code to a specific provider. When you want to add a new provider or run tests without real API calls, the coupling makes it hard. Define a provider interface instead:
// Define the interface
interface LLMProvider {
complete(messages: ChatMessage[], options?: ChatOptions): Promise<ChatResponse>;
stream(messages: ChatMessage[], options?: ChatOptions): AsyncGenerator<StreamEvent>;
}
// Provider implementations conform to the interface
class OpenAIProvider implements LLMProvider { ... }
class AnthropicProvider implements LLMProvider { ... }
class MockProvider implements LLMProvider { ... } // For tests
// Inject via dependency injection
function buildChatService(provider: LLMProvider) {
return { chat: (msg: string) => provider.complete([{ role: "user", content: msg }]) };
}
Best Practice: Use Environment-Based Provider Selection
Switch providers via environment variables without code changes:
function createProvider(): LLMProvider {
switch (process.env.LLM_PROVIDER) {
case "anthropic": return new AnthropicProvider();
case "mock": return new MockProvider(); // For CI/testing
default: return new OpenAIProvider();
}
}
Named Product Reference: Vercel AI SDK Provider Abstraction
The Vercel AI SDK already implements this pattern — @ai-sdk/openai, @ai-sdk/anthropic, and @ai-sdk/google all conform to the same interface. Switching providers requires changing one import. For teams that want provider flexibility without writing the abstraction layer themselves, starting with the Vercel AI SDK is the right choice.
Key Takeaways
- •Vercel AI SDK provides a unified interface across AI providers
- •generateObject returns typed objects validated against Zod schemas
- •Switch providers with one line — same interface everywhere
- •Full-stack pattern: AI SDK server + Zod shared + React hooks client
Progress is kept in this browser only.