vercel-ai-sdk-knowledge-patch

Vercel AI SDK Knowledge Patch

Claude Opus 4.6 knows AI SDK through 3.x. It is unaware of the features below, which cover AI SDK 4.0 (2024-11-18) through 4.1 (2025-01-20).

Index

Topic	Reference	Key features
4.0 & 4.1 features	references/ai-sdk-4-features.md	PDF files, computer use tools, continuation, image generation, stream smoothing, createDataStreamResponse, tool context/repair, structured outputs with tools, new providers

---

PDF Support (4.0)

Send PDFs as file type in message content:

const result = await generateText({
  model: anthropic('claude-3-5-sonnet-20241022'),
  messages: [{
    role: 'user',
    content: [
      { type: 'text', text: 'Summarize this.' },
      { type: 'file', data: fs.readFileSync('./doc.pdf'), mimeType: 'application/pdf' },
    ],
  }],
});

Works with Anthropic, Google Generative AI, and Vertex AI.

---

Computer Use (4.0)

Anthropic tools accessible via anthropic.tools:

const computerTool = anthropic.tools.computer_20241022({
  displayWidthPx: 1920, displayHeightPx: 1080,
  execute: async ({ action, coordinate, text }) => { /* implement actions */ },
  experimental_toToolResultContent: (result) => /* format result */,
});

await generateText({
  model: anthropic('claude-3-5-sonnet-20241022'),
  tools: { computer: computerTool },
  maxSteps: 10,
});

Also: anthropic.tools.textEditor_20241022(), anthropic.tools.bash_20241022().

---

Continuation (4.0)

Auto-continue when generation hits length limit:

await generateText({
  model: openai('gpt-4o'),
  maxSteps: 5,
  experimental_continueSteps: true,
  prompt: 'Write a long essay...',
});

---

Image Generation (4.1)

import { experimental_generateImage as generateImage } from 'ai';
import { replicate } from '@ai-sdk/replicate';

const { image } = await generateImage({
  model: replicate.image('black-forest-labs/flux-1.1-pro-ultra'),
  prompt: 'A cityscape at sunset',
  size: '16:9',
  n: 3,
});
// image.base64, image.uint8Array

Providers: replicate.image(), openai.image(), vertex.image(), fireworks.image().

---

Stream Smoothing (4.1)

import { smoothStream, streamText } from 'ai';

const result = streamText({
  model, prompt,
  experimental_transform: smoothStream(),
});

---

Non-blocking Data Streaming (4.1)

Stream custom data before/alongside LLM output:

import { createDataStreamResponse, streamText } from 'ai';

return createDataStreamResponse({
  execute: async (dataStream) => {
    dataStream.writeData({ type: 'source', url: '...' });
    const result = streamText({ model, messages });
    result.mergeIntoDataStream(dataStream);
  },
});

---

Tool Improvements (4.1)

Execute context: execute(args, { toolCallId, messages, abortSignal }).

Tool call repair:

await generateText({
  model, tools, prompt,
  experimental_repairToolCall: async ({ toolCall, tools, parameterSchema, error }) => {
    if (NoSuchToolError.isInstance(error)) return null;
    const { object } = await generateObject({ model, schema: tools[toolCall.toolName].parameters, prompt: '...' });
    return { ...toolCall, args: JSON.stringify(object) };
  },
});

Structured outputs with tools via experimental_output: Output.object({ schema }) (OpenAI only).

---

New Providers

Package	Provider
@ai-sdk/xai	xAI Grok
@ai-sdk/groq	Groq
@ai-sdk/replicate	Replicate (image)
@ai-sdk/fireworks	Fireworks (language + image)
@ai-sdk/deepinfra	DeepInfra
@ai-sdk/deepseek	DeepSeek
@ai-sdk/cerebras	Cerebras

---

Reference Files

File	Contents
ai-sdk-4-features.md	Complete API reference for all 4.0 and 4.1 features with full code examples