Migrate AI SDK 3.4 to 4.0

Recommended Migration Process

1. Backup your project. If you use a versioning control system, make sure all previous versions are committed.
2. Migrate to AI SDK 3.4.
3. Upgrade to AI SDK 4.0.
4. Automatically migrate your code using codemods.

   If you don't want to use codemods, we recommend resolving all deprecation warnings before upgrading to AI SDK 4.0.

5. Follow the breaking changes guide below.
6. Verify your project is working as expected.
7. Commit your changes.

AI SDK 4.0 package versions

You need to update the following packages to the following versions in your package.json file(s):

• ai package: 4.0.*
• ai-sdk@provider-utils package: 2.0.*
• ai-sdk/* packages: 1.0.* (other @ai-sdk packages)

Codemods

The AI SDK provides Codemod transformations to help upgrade your codebase when a feature is deprecated, removed, or otherwise changed.

Codemods are transformations that run on your codebase programmatically. They allow you to easily apply many changes without having to manually go through every file.

You can run all codemods provided as part of the 4.0 upgrade process by running the following command from the root of your project:

npx @ai-sdk/codemod upgrade

Individual codemods can be run by specifying the name of the codemod:

npx @ai-sdk/codemod <codemod-name> <path>

See also the table of codemods. In addition, the latest set of codemods can be found in the @ai-sdk/codemod repository.

Provider Changes

Removed baseUrl option

The baseUrl option has been removed from all providers. Please use the baseURL option instead.

const perplexity = createOpenAI({
  // ...
  baseUrl: 'https://api.perplexity.ai/',
});

const perplexity = createOpenAI({
  // ...
  baseURL: 'https://api.perplexity.ai/',
});

Anthropic Provider

Removed Anthropic facade

The Anthropic facade has been removed from the Anthropic provider. Please use the anthropic object or the createAnthropic function instead.

const anthropic = new Anthropic({
  // ...
});

const anthropic = createAnthropic({
  // ...
});

Removed topK setting

The model specific topK setting has been removed from the Anthropic provider. You can use the standard topK setting instead.

const result = await generateText({
  model: anthropic('claude-3-5-sonnet-latest', {
    topK: 0.5,
  }),
});

const result = await generateText({
  model: anthropic('claude-3-5-sonnet-latest'),
  topK: 0.5,
});

Google Generative AI Provider

Removed Google facade

The Google facade has been removed from the Google Generative AI provider. Please use the google object or the createGoogleGenerativeAI function instead.

const google = new Google({
  // ...
});

const google = createGoogleGenerativeAI({
  // ...
});

Removed topK setting

The model-specific topK setting has been removed from the Google Generative AI provider. You can use the standard topK setting instead.

const result = await generateText({
  model: google('gemini-1.5-flash', {
    topK: 0.5,
  }),
});

const result = await generateText({
  model: google('gemini-1.5-flash'),
  topK: 0.5,
});

Google Vertex Provider

Removed topK setting

The model-specific topK setting has been removed from the Google Vertex provider. You can use the standard topK setting instead.

const result = await generateText({
  model: vertex('gemini-1.5-flash', {
    topK: 0.5,
  }),
});

const result = await generateText({
  model: vertex('gemini-1.5-flash'),
  topK: 0.5,
});

Mistral Provider

Removed Mistral facade

The Mistral facade has been removed from the Mistral provider. Please use the mistral object or the createMistral function instead.

const mistral = new Mistral({
  // ...
});

const mistral = createMistral({
  // ...
});

OpenAI Provider

Removed OpenAI facade

The OpenAI facade has been removed from the OpenAI provider. Please use the openai object or the createOpenAI function instead.

const openai = new OpenAI({
  // ...
});

const openai = createOpenAI({
  // ...
});

LangChain Adapter

Removed toAIStream

The toAIStream function has been removed from the LangChain adapter. Please use the toDataStream function instead.

LangChainAdapter.toAIStream(stream);

LangChainAdapter.toDataStream(stream);

AI SDK Core Changes

streamText returns immediately

Instead of returning a Promise, the streamText function now returns immediately. It is not necessary to await the result of streamText.

const result = await streamText({
  // ...
});

const result = streamText({
  // ...
});

streamObject returns immediately

Instead of returning a Promise, the streamObject function now returns immediately. It is not necessary to await the result of streamObject.

const result = await streamObject({
  // ...
});

const result = streamObject({
  // ...
});

Remove roundtrips

The maxToolRoundtrips and maxAutomaticRoundtrips options have been removed from the generateText and streamText functions. Please use the maxSteps option instead.

The roundtrips property has been removed from the GenerateTextResult type. Please use the steps property instead.

const { text, roundtrips } = await generateText({
  maxToolRoundtrips: 1, // or maxAutomaticRoundtrips
  // ...
});

const { text, steps } = await generateText({
  maxSteps: 2,
  // ...
});

Removed nanoid export

The nanoid export has been removed. Please use generateId instead.

import { nanoid } from 'ai';

import { generateId } from 'ai';

Increased default size of generated IDs

The generateId function now generates 16-character IDs. The previous default was 7 characters.

This might e.g. require updating your database schema if you limit the length of IDs.

import { generateId } from 'ai';

const id = generateId(); // now 16 characters

Removed ExperimentalMessage types

The following types have been removed:

• ExperimentalMessage (use CoreMessage instead)
• ExperimentalUserMessage (use CoreUserMessage instead)
• ExperimentalAssistantMessage (use CoreAssistantMessage instead)
• ExperimentalToolMessage (use CoreToolMessage instead)

import {
  ExperimentalMessage,
  ExperimentalUserMessage,
  ExperimentalAssistantMessage,
  ExperimentalToolMessage,
} from 'ai';

import {
  CoreMessage,
  CoreUserMessage,
  CoreAssistantMessage,
  CoreToolMessage,
} from 'ai';

Removed ExperimentalTool type

The ExperimentalTool type has been removed. Please use the CoreTool type instead.

import { ExperimentalTool } from 'ai';

import { CoreTool } from 'ai';

Removed experimental AI function exports

The following exports have been removed:

• experimental_generateText (use generateText instead)
• experimental_streamText (use streamText instead)
• experimental_generateObject (use generateObject instead)
• experimental_streamObject (use streamObject instead)

import {
  experimental_generateText,
  experimental_streamText,
  experimental_generateObject,
  experimental_streamObject,
} from 'ai';

import { generateText, streamText, generateObject, streamObject } from 'ai';

Removed AI-stream related methods from streamText

The following methods have been removed from the streamText result:

• toAIStream
• pipeAIStreamToResponse
• toAIStreamResponse

Use the toDataStream, pipeDataStreamToResponse, and toDataStreamResponse functions instead.

const result = await streamText({
  // ...
});

result.toAIStream();
result.pipeAIStreamToResponse(response);
result.toAIStreamResponse();

const result = streamText({
  // ...
});

result.toDataStream();
result.pipeDataStreamToResponse(response);
result.toDataStreamResponse();

Renamed "formatStreamPart" to "formatDataStreamPart"

The formatStreamPart function has been renamed to formatDataStreamPart.

formatStreamPart('text', 'Hello, world!');

formatDataStreamPart('text', 'Hello, world!');

Renamed "parseStreamPart" to "parseDataStreamPart"

The parseStreamPart function has been renamed to parseDataStreamPart.

const part = parseStreamPart(line);

const part = parseDataStreamPart(line);

Renamed TokenUsage, CompletionTokenUsage and EmbeddingTokenUsage types

The TokenUsage, CompletionTokenUsage and EmbeddingTokenUsage types have been renamed to LanguageModelUsage (for the first two) and EmbeddingModelUsage (for the last).

import { TokenUsage, CompletionTokenUsage, EmbeddingTokenUsage } from 'ai';

import { LanguageModelUsage, EmbeddingModelUsage } from 'ai';

Removed deprecated telemetry data

The following telemetry data values have been removed:

• ai.finishReason (now in ai.response.finishReason)
• ai.result.object (now in ai.response.object)
• ai.result.text (now in ai.response.text)
• ai.result.toolCalls (now in ai.response.toolCalls)
• ai.stream.msToFirstChunk (now in ai.response.msToFirstChunk)

This change will apply to observability providers and any scripts or automation that you use for processing telemetry data.

Provider Registry

Removed experimental_Provider, experimental_ProviderRegistry, and experimental_ModelRegistry

The experimental_Provider interface, experimental_ProviderRegistry interface, and experimental_ModelRegistry interface have been removed. Please use the Provider interface instead.

import { experimental_Provider, experimental_ProviderRegistry } from 'ai';

import { Provider } from 'ai';

Removed experimental_​createModelRegistry function

The experimental_createModelRegistry function has been removed. Please use the experimental_createProviderRegistry function instead.

import { experimental_createModelRegistry } from 'ai';

import { experimental_createProviderRegistry } from 'ai';

Removed rawResponse from results

The rawResponse property has been removed from the generateText, streamText, generateObject, and streamObject results. You can use the response property instead.

const { text, rawResponse } = await generateText({
  // ...
});

const { text, response } = await generateText({
  // ...
});

Removed init option from pipeDataStreamToResponse and toDataStreamResponse

The init option has been removed from the pipeDataStreamToResponse and toDataStreamResponse functions. You can set the values from init directly into the options object.

const result = await streamText({
  // ...
});

result.toDataStreamResponse(response, {
  init: {
    headers: {
      'X-Custom-Header': 'value',
    },
  },
  // ...
});

const result = streamText({
  // ...
});

result.toDataStreamResponse(response, {
  headers: {
    'X-Custom-Header': 'value',
  },
  // ...
});

Removed responseMessages from generateText and streamText

The responseMessages property has been removed from the generateText and streamText results. This includes the onFinish callback. Please use the response.messages property instead.

const { text, responseMessages } = await generateText({
  // ...
});

const { text, response } = await generateText({
  // ...
});

const responseMessages = response.messages;

Removed experimental_​continuationSteps option

The experimental_continuationSteps option has been removed from the generateText function. Please use the experimental_continueSteps option instead.

const result = await generateText({
  experimental_continuationSteps: true,
  // ...
});

const result = await generateText({
  experimental_continueSteps: true,
  // ...
});

Removed LanguageModelResponseMetadataWithHeaders type

The LanguageModelResponseMetadataWithHeaders type has been removed. Please use the LanguageModelResponseMetadata type instead.

import { LanguageModelResponseMetadataWithHeaders } from 'ai';

import { LanguageModelResponseMetadata } from 'ai';

Changed streamText warnings result to Promise

The warnings property of the StreamTextResult type is now a Promise.

const result = await streamText({
  // ...
});

const warnings = result.warnings;

const result = streamText({
  // ...
});

const warnings = await result.warnings;

Changed streamObject warnings result to Promise

The warnings property of the StreamObjectResult type is now a Promise.

const result = await streamObject({
  // ...
});

const warnings = result.warnings;

const result = streamObject({
  // ...
});

const warnings = await result.warnings;

Renamed simulateReadableStream values to chunks

The simulateReadableStream function from ai/test has been renamed to chunks.

import { simulateReadableStream } from 'ai/test';

const stream = simulateReadableStream({
  values: [1, 2, 3],
  chunkDelayInMs: 100,
});

import { simulateReadableStream } from 'ai/test';

const stream = simulateReadableStream({
  chunks: [1, 2, 3],
  chunkDelayInMs: 100,
});

AI SDK RSC Changes

Removed render function

The AI SDK RSC 3.0 render function has been removed. Please use the streamUI function instead or switch to AI SDK UI.

import { render } from 'ai/rsc';

import { streamUI } from 'ai/rsc';

AI SDK UI Changes

Removed Svelte, Vue, and SolidJS exports

The ai package no longer exports Svelte, Vue, and SolidJS UI integrations. You need to install the @ai-sdk/svelte, @ai-sdk/vue, and @ai-sdk/solid packages directly.

import { useChat } from 'ai/svelte';

import { useChat } from '@ai-sdk/svelte';

Removed experimental_StreamData

The experimental_StreamData export has been removed. Please use the StreamData export instead.

import { experimental_StreamData } from 'ai';

import { StreamData } from 'ai';

useChat hook

Removed streamMode setting

The streamMode options has been removed from the useChat hook. Please use the streamProtocol parameter instead.

const { messages } = useChat({
  streamMode: 'text',
  // ...
});

const { messages } = useChat({
  streamProtocol: 'text',
  // ...
});

Replaced roundtrip setting with maxSteps

The following options have been removed from the useChat hook:

• experimental_maxAutomaticRoundtrips
• maxAutomaticRoundtrips
• maxToolRoundtrips

Please use the maxSteps option instead. The value of maxSteps is equal to roundtrips + 1.

const { messages } = useChat({
  experimental_maxAutomaticRoundtrips: 2,
  // or maxAutomaticRoundtrips
  // or maxToolRoundtrips
  // ...
});

const { messages } = useChat({
  maxSteps: 3, // 2 roundtrips + 1
  // ...
});

Removed options setting

The options parameter in the useChat hook has been removed. Please use the headers and body parameters instead.

const { messages } = useChat({
  options: {
    headers: {
      'X-Custom-Header': 'value',
    },
  },
  // ...
});

const { messages } = useChat({
  headers: {
    'X-Custom-Header': 'value',
  },
  // ...
});

Removed experimental_addToolResult method

The experimental_addToolResult method has been removed from the useChat hook. Please use the addToolResult method instead.

const { messages, experimental_addToolResult } = useChat({
  // ...
});

const { messages, addToolResult } = useChat({
  // ...
});

Changed default value of keepLastMessageOnError to true and deprecated the option

The keepLastMessageOnError option has been changed to default to true. The option will be removed in the next major release.

const { messages } = useChat({
  keepLastMessageOnError: true,
  // ...
});

const { messages } = useChat({
  // ...
});

useCompletion hook

Removed streamMode setting

The streamMode options has been removed from the useCompletion hook. Please use the streamProtocol parameter instead.

const { text } = useCompletion({
  streamMode: 'text',
  // ...
});

const { text } = useCompletion({
  streamProtocol: 'text',
  // ...
});

useAssistant hook

Removed experimental_useAssistant export

The experimental_useAssistant export has been removed from the useAssistant hook. Please use the useAssistant hook directly instead.

import { experimental_useAssistant } from 'ai/react';

import { useAssistant } from 'ai/react';

Removed threadId and messageId from AssistantResponse

The threadId and messageId parameters have been removed from the AssistantResponse function. Please use the threadId and messageId variables from the outer scope instead.

return AssistantResponse(
  { threadId: myThreadId, messageId: myMessageId },
  async ({ forwardStream, sendDataMessage, threadId, messageId }) => {
    // use threadId and messageId here
  },
);

return AssistantResponse(
  { threadId: myThreadId, messageId: myMessageId },
  async ({ forwardStream, sendDataMessage }) => {
    // use myThreadId and myMessageId here
  },
);

Removed experimental_​AssistantResponse export

The experimental_AssistantResponse export has been removed. Please use the AssistantResponse function directly instead.

import { experimental_AssistantResponse } from 'ai';

import { AssistantResponse } from 'ai';

experimental_useObject hook

The setInput helper has been removed from the experimental_useObject hook. Please use the submit helper instead.

const { object, setInput } = useObject({
  // ...
});

const { object, submit } = useObject({
  // ...
});

AI SDK Errors

Removed isXXXError static methods

The isXXXError static methods have been removed from AI SDK errors. Please use the isInstance method of the corresponding error class instead.

import { APICallError } from 'ai';

APICallError.isAPICallError(error);

import { APICallError } from 'ai';

APICallError.isInstance(error);

Removed toJSON method

The toJSON method has been removed from AI SDK errors.

AI SDK 2.x Legacy Changes

Removed 2.x legacy providers

Legacy providers from AI SDK 2.x have been removed. Please use the new AI SDK provider architecture instead.

Removed 2.x legacy function and tool calling

The legacy function_call and tools options have been removed from useChat and Message. The name property from the Message type has been removed. Please use the AI SDK Core tool calling instead.

Removed 2.x prompt helpers

Prompt helpers for constructing message prompts are no longer needed with the AI SDK provider architecture and have been removed.

Removed 2.x AIStream

The AIStream function and related exports have been removed. Please use the streamText function and its toDataStream() method instead.

Removed 2.x StreamingTextResponse

The StreamingTextResponse function has been removed. Please use the streamText function and its toDataStreamResponse() method instead.

Removed 2.x streamToResponse

The streamToResponse function has been removed. Please use the streamText function and its pipeDataStreamToResponse() method instead.

Removed 2.x RSC Tokens streaming

The legacy Tokens RSC streaming from 2.x has been removed. Tokens were implemented prior to AI SDK RSC and are no longer needed.

Codemod Table

The following table lists codemod availability for the AI SDK 4.0 upgrade process. Note the codemod upgrade command will run all of them for you. This list is provided to give visibility into which migrations have some automation. It can also be helpful to find the codemod names if you'd like to run a subset of codemods. For more, see the Codemods section.