Skip to main content
The Vercel AI SDK provides powerful functions for text generation and streaming. Below are details and examples for implementing these functions.

​
Text Generation

Use generateText() for standard, non-streaming responses. 
import { generateText } from 'ai';
import { openai, MODEL } from './client';

async function generateGreeting() {
  const response = await generateText({
    model: openai(MODEL),
    prompt: 'Explain quantum computing in one sentence.',
  });

  console.log(response.text);
}

​
Streaming Responses

Use streamText() to stream tokens as they are generated. 
import { streamText } from 'ai';
import { openai, MODEL } from './client';

async function streamResponse() {
  const result = await streamText({
    model: openai(MODEL),
    prompt: 'Write a poem about a software engineer.',
  });

  for await (const chunk of result.textStream) {
    process.stdout.write(chunk);
  }
}
Streaming improves perceived latency and is recommended for chat interfaces.

​
System Prompts & Parameters

You can customize model behavior using a system prompt and generation parameters.

​
Supported Parameters

​
model
LanguageModel
required
The model instance to use for generation.
​
prompt
string
The text prompt to generate a response for. Either prompt or messages is required.
​
system
string
System instructions to guide the model’s behavior.
​
temperature
number
Controls randomness. Lower values produce more deterministic responses.
​
maxTokens
number
Maximum number of tokens generated.
​
topP
number
Nucleus sampling probability.
​
topK
number
Limits sampling to the top K probable tokens.
​
presencePenalty
number
Encourages introducing new topics.
​
frequencyPenalty
number
Reduces repeated words or phrases.
​
seed
number
Produces deterministic outputs when supported.
​
stopSequences
string[]
Stops generation when one of the specified sequences is encountered.

​
Error Handling

Wrap SDK calls in a try...catch block to handle API errors, validation failures, and authentication issues. 
import { generateText } from 'ai';
import { openai, MODEL } from './client';

async function safeGenerate() {
  try {
    const response = await generateText({
      model: openai(MODEL),
      prompt: 'Hello!',
      temperature: -1,
    });

    console.log(response.text);
  } catch (error: any) {
    console.error('Error:', error.name);
    console.error(error.message);
  }
}
Common errors include:
  • Invalid API key
  • Incorrect endpoint URL
  • Unsupported model
  • Invalid request parameters
  • Rate limits