Skip to main content
The Vercel AI SDK provides a standardized way to interact with multiple providers. Simply wrap your model with observ.wrap() and everything works automatically.

​
Why Use Vercel AI SDK?

Provider Flexibility

Switch between OpenAI, Anthropic, Google, Mistral, and 20+ other providers without changing your code

Full Feature Support

Streaming, structured outputs (generateObject), and tool calling all work automatically

Automatic Caching

Enable recall: true for intelligent semantic caching across all providers

Minimal Setup

Wrap your model with observ.wrap() and use it normally - that’s it

​
Quick Example

import { Observ } from "observ-sdk";
import { openai } from "@ai-sdk/openai";
import { generateText } from "ai";

const observ = new Observ({
  apiKey: "your-observ-api-key",
  recall: true, // Enable semantic caching
});

// Wrap the model
const model = observ.wrap(openai("gpt-4"));

// Use it normally
const result = await generateText({
  model,
  prompt: "What is TypeScript?",
});

console.log(result.text);

​
Supported Providers

The Vercel AI SDK supports 25+ providers. Here are some popular ones:
  • OpenAI - GPT-4, GPT-3.5
  • Anthropic - Claude Sonnet 4.5, Claude Opus
  • Google - Gemini Pro, Gemini Flash
  • Mistral - Mistral Large, Mistral Medium
  • Cohere - Command R+
  • And many more…

​
Features

​
Streaming

Streaming works automatically with full observability: 
import { streamText } from "ai";

const stream = await streamText({
  model, // Your wrapped model
  prompt: "Write a detailed explanation of async/await",
});

for await (const chunk of stream.textStream) {
  process.stdout.write(chunk);
}

​
Structured Outputs

Generate typed objects with generateObject: 
import { generateObject } from "ai";
import { z } from "zod";

const result = await generateObject({
  model,
  schema: z.object({
    name: z.string(),
    age: z.number(),
  }),
  prompt: "Generate a person's profile",
});

​
Tool Calling

Use tools (function calling) with any compatible provider: 
import { generateText, tool } from "ai";
import { z } from "zod";

const result = await generateText({
  model,
  tools: {
    weather: tool({
      description: "Get weather information",
      parameters: z.object({
        location: z.string(),
      }),
      execute: async ({ location }) => ({
        temperature: 72,
        location,
      }),
    }),
  },
  prompt: "What's the weather in San Francisco?",
});

​
Next Steps

Installation

Install the SDK and dependencies

Usage Guide

Detailed usage with all providers

Sessions

Track conversation threads

Metadata

Add custom metadata to traces