ai

Patch Changes

• Updated dependencies [35f93ce]
  • @ai-sdk/gateway@1.0.4

Patch Changes

• 8e72304: fix (ai): handle invalid tool calls

Patch Changes

• d983eee: feat(ai): allow passing model string for embeddings

Patch Changes

• Updated dependencies [444df49]
  • @ai-sdk/gateway@1.0.2

Patch Changes

• 90d212f: feat (ai): add experimental tool call context
• Updated dependencies
  • @ai-sdk/gateway@1.0.1
  • @ai-sdk/provider-utils@3.0.1

Patch Changes

• 401d73e: fix (ai): support dynamic tool calls in lastAssistantMessageIsCompleteWithToolCalls
• 69fde99: Set status to ready when reconnect stream is null

Patch Changes

• 4d0c108: fix(ai/ui): convert provider metadata for system messages to model messages

Major Changes

• 38fe0e5: BREAKING: Migrate language generation from the Sonar Chat Completions API to the Agent API. Replace Sonar model IDs and provider options with Agent API presets, models, and tools. The new API changes request and response metadata, raw stream events, usage and cost data, and does not support Sonar PDF input or image and video results.

Patch Changes

• 38fe0e5: Recover missing text from Agent API terminal events and output items, including incomplete responses, without repeating text already received as deltas. Track each message content part separately.
• 38fe0e5: Accept native Agent API tool traces such as finance results without validating them as web search results. Preserve these traces in raw responses and stream chunks.
• 38fe0e5: Preserve URL citation annotations as sources in Agent API streams, including completed and incomplete terminal responses.
• 38fe0e5: Emit each Agent API source URL once while preserving search result IDs for citation correlation when a URL is fetched or annotated before it appears in search results.

Major Changes

• e1cbf8a: chore(@ai-sdk/rsc): extract to separate package

• a847c3e: chore: rename reasoning to reasoningText etc

• 13fef90: chore (ai): remove automatic conversion of UI messages to model messages

• d964901: - remove setting temperature to 0 by default

  • remove null option from DefaultSettingsMiddleware
  • remove setting defaults for temperature and stopSequences in ai to enable middleware changes

• 0a710d8: feat (ui): typed tool parts in ui messages

• 9ad0484: feat (ai): automatic tool execution error handling

• 63f9e9b: chore (provider,ai): tools have input/output instead of args,result

• ab7ccef: chore (ai): change source ui message parts to source-url

• d5f588f: AI SDK 5

• ec78cdc: chore (ai): remove "data" UIMessage role

• 6a83f7d: refactoring (ai): restructure message metadata transfer

• db345da: chore (ai): remove exports of internal ui functions

• 496bbc1: chore (ui): inline/remove ChatRequest type

• 72d7d72: chore (ai): stable activeTools

• 40acf9b: feat (ui): introduce ChatStore and ChatTransport

• 98f25e5: chore (ui): remove managed chat inputs

• 2d03e19: chore (ai): remove StreamCallbacks.onCompletion

• da70d79: chore (ai): remove getUIText helper

• c60f895: chore (ai): remove useChat keepLastMessageOnError

• 0560977: chore (ai): improve consistency of generate text result, stream text result, and step result

• 9477ebb: chore (ui): remove useAssistant hook (breaking change)

• 1f55c21: chore (ai): send reasoning to the client by default

• e7dc6c7: chore (ai): remove onResponse callback

• 8b86e99: chore (ai): replace Message with UIMessage

• 04d5063: chore (ai): rename default provider global to AI_SDK_DEFAULT_PROVIDER

• 319b989: chore (ai): remove content from ui messages

• 14c9410: chore: refactor file towards source pattern (spec)

• a34eb39: chore (ai): remove data and allowEmptySubmit from ChatRequestOptions

• f04fb4a: chore (ai): replace useChat attachments with file ui parts

• f7e8bf4: chore (ai): flatten ui message stream parts

• 257224b: chore (ai): separate TextStreamChatTransport

• fd1924b: chore (ai): remove redundant mimeType property

• 2524fc7: chore (ai): remove ui message toolInvocations property

• 6fba4c7: chore (ai): remove deprecated experimental_providerMetadata

• b4b4bb2: chore (ui): rename experimental_resume to resumeStream

• 441d042: chore (ui): data stream protocol v2 with SSEs

• ef256ed: chore (ai): refactor and use chatstore in svelte

• 516be5b: ### Move Image Model Settings into generate options

  Image Models no longer have settings. Instead, maxImagesPerCall can be passed directly to generateImage(). All other image settings can be passed to providerOptions[provider].

  Before

  await generateImage({
    model: luma.image("photon-flash-1", {
      maxImagesPerCall: 5,
      pollIntervalMillis: 500,
    }),
    prompt,
    n: 10,
  });
  

  After

  await generateImage({
    model: luma.image("photon-flash-1"),
    prompt,
    n: 10,
    maxImagesPerCall: 5,
    providerOptions: {
      luma: { pollIntervalMillis: 5 },
    },
  });
  

  Pull Request: https://github.com/vercel/ai/pull/6180

• a662dea: chore (ai): remove sendExtraMessageFields

• d884051: feat (ai): simplify default provider setup

• e8324c5: feat (ai): add args callbacks to tools

• fafc3f2: chore (ai): change file to parts to use urls instead of data

• 1ed0287: chore (ai): stable sendStart/sendFinish options

• c7710a9: chore (ai): rename DataStreamToSSETransformStream to JsonToSseTransformStream

• bfbfc4c: feat (ai): streamText/generateText: totalUsage contains usage for all steps. usage is for a single step.

• 9ae327d: chore (ui): replace chat store concept with chat instances

• 9315076: chore (ai): rename continueUntil to stopWhen. Rename maxSteps stop condition to isStepCount.

• 247ee0c: chore (ai): remove steps from tool invocation ui parts

• 109c0ac: chore (ai): rename id to chatId (in post request, resume request, and useChat)

• 954aa73: feat (ui): extended regenerate support

• 33eb499: feat (ai): inject message id in createUIMessageStream

• 901df02: feat (ui): use UI_MESSAGE generic

• 4892798: chore (ai): always stream tool calls

• c25cbce: feat (ai): use console.error as default error handler for streamText and streamObject

• b33ed7a: chore (ai): rename DataStream* to UIMessage*

• ed675de: feat (ai): add ui data parts

• 7bb58d4: chore (ai): restructure prepareRequest

• ea7a7c9: feat (ui): UI message metadata

• 0463011: fix (ai): update source url stream part

• dcc549b: remove StreamTextResult.mergeIntoDataStream method rename DataStreamOptions.getErrorMessage to onError add pipeTextStreamToResponse function add createTextStreamResponse function change createDataStreamResponse function to accept a DataStream and not a DataStreamWriter change pipeDataStreamToResponse function to accept a DataStream and not a DataStreamWriter change pipeDataStreamToResponse function to have a single parameter

• 35fc02c: chore (ui): rename RequestOptions to CompletionRequestOptions

• 64f6d64: feat (ai): replace maxSteps with continueUntil (generateText)

• 175b868: chore (ai): rename reasoning UI parts 'reasoning' property to 'text'

• 60e2c56: feat (ai): restructure chat transports

• 765f1cd: chore (ai): remove deprecated useChat isLoading helper

• cb2b53a: chore (ai): refactor header preparation

• e244a78: chore (ai): remove StreamData and mergeStreams

• d306260: feat (ai): replace maxSteps with continueUntil (streamText)

• 4bfe9ec: chore (ai): remove ui message reasoning property

• 1766ede: chore: rename maxTokens to maxOutputTokens

• 2877a74: chore (ai): remove ui message data property

• 1409e13: chore (ai): remove experimental continueSteps

• b32e192: chore (ai): rename reasoning to reasoningText, rename reasoningDetails to reasoning (streamText, generateText)

• 92cb0a2: chore (ai): rename CoreMessage to ModelMessage

• 2b637d6: chore (ai): rename UIMessageStreamPart to UIMessageChunk

Minor Changes

• b7eae2d: feat (core): Add finishReason field to NoObjectGeneratedError
• bcea599: feat (ai): add content to generateText result
• 48d675a: feat (ai): add content to streamText result
• c9ad635: feat (ai): add filename to file ui parts

Patch Changes

• a571d6e: chore(provider-utils): move ToolResultContent to provider-utils

• de2d2ab: feat(ai): add provider and provider registry middleware functionality

• c22ad54: feat(smooth-stream): chunking callbacks

• d88455d: feat (ai): expose http chat transport type

• e7fcc86: feat (ai): introduce dynamic tools

• da1e6f0: feat (ui): add generics to ui message stream parts

• 48378b9: fix (ai): send null as tool output when tools return undefined

• 5d1e3ba: chore (ai): remove provider re-exports

• 93d53a1: chore (ai): remove cli

• e90d45d: chore (rsc): move HANGING_STREAM_WARNING_TIME constant into @ai-sdk/rsc package

• b32c141: feat (ai): add array support to stopWhen

• bc3109f: chore (ai): push stream-callbacks into langchain/llamaindex adapters

• 0d9583c: fix (ai): use user-provided media type when available

• 38ae5cc: feat (ai): export InferUIMessageChunk type

• 10b21eb: feat(cli): add ai command line interface

• 9e40cbe: Allow destructuring output and errorText on ToolUIPart type

• 6909543: feat (ai): support system parameter in Agent constructor

• 86cfc72: feat (ai): add ignoreIncompleteToolCalls option to convertToModelMessages

• 377bbcf: fix (ui): tool input can be undefined during input-streaming

• d8aeaef: feat(providers/fal): add transcribe

• ae77a99: chore (ai): rename text and reasoning chunks in streamText fullstream

• 4fef487: feat: support for zod v4 for schema validation

  All these methods now accept both a zod v4 and zod v3 schemas for validation:

  • generateObject()
  • streamObject()
  • generateText()
  • experimental_useObject() from @ai-sdk/react
  • streamUI() from @ai-sdk/rsc

• b1e3abd: feat (ai): expose ui message stream headers

• 4f3e637: fix (ui): avoid caching globalThis.fetch in case it is patched by other libraries

• 14cb3be: chore(providers/llamaindex): extract to separate package

• 1f6ce57: feat (ai): infer tool call types in the onToolCall callback

• 16ccfb2: feat (ai): add readUIMessageStream helper

• 225f087: fix (ai/mcp): prevent mutation of customEnv

• ce1d1f3: feat (ai): export mock image, speech, and transcription models

• fc0380b: feat (ui): resolvable header, body, credentials in http chat transport

• 6622441: feat (ai): add static/dynamic toolCalls/toolResults helpers

• 4048ce3: fix (ai): add tests and examples for openai responses

• 6c42e56: feat (ai): validate ui stream data chunks

• bedb239: chore (ai): make ui stream parts value optional when it's not required

• 9b4d074: feat(streamObject): add enum support

• c8fce91: feat (ai): add experimental Agent abstraction

• 655cf3c: feat (ui): add onFinish to createUIMessageStream

• 3e10408: fix(utils/detect-mimetype): add support for detecting id3 tags

• d5ae088: feat (ui): add sendAutomaticallyWhen to Chat

• ced8eee: feat(ai): re-export zodSchema from main package

• c040e2f: fix (ui): inject generated response message id

• d3960e3: selectTelemetryAttributes more robustness

• faea29f: fix (provider/openai): multi-step reasoning with text

• 66af894: fix (ai): respect content order in toResponseMessages

• 332167b: chore (ai): move maxSteps into UseChatOptions

• 6b1c55c: feat (ai): introduce GLOBAL_DEFAULT_PROVIDER

• 5a975a4: feat (ui): update Chat tool result submission

• 507ac1d: fix (ui/react): update messages immediately with the submitted user message

• a166433: feat: add transcription with experimental_transcribe

• 26735b5: chore(embedding-model): add v2 interface

• c93a8bc: chore(ai): export AsyncIterableStream type from async-iterable-stream module

• 0d2c085: feat (ai): support string model ids through gateway

• 2b9bbcd: feat (ai): improve prompt validation error message

• a8c8bd5: feat(embed-many): respect supportsParallelCalls & concurrency

• 75c3396: fix (ai): handle errors in 2nd streamText doStream call

• cb9c9e4: remove deprecated experimental_wrapLanguageModel

• 9bf7291: chore(providers/openai): enable structuredOutputs by default & switch to provider option

• 9b0da33: fix (ai): do not send id with start unless specified

• 28ad69e: fix(react-native): support experimental_attachments without FileList global

• 0b78e17: chore(ai/generateObject): simplify function signature

• 20398f2: feat: ai sdk cli documentation + adjusted default model

• 66962ed: fix(packages): export node10 compatible types

• b71fe8d: fix(ai): remove jsondiffpatch dependency

• 7827a49: fix (ai/core): refactor toResponseMessages to filter out empty string/content

• bd8a36c: feat(embedding-model-v2/embedMany): add response body field

• d9209ca: fix (image-model): specificationVersion: v1 -> v2

• b346545: feat (ai): add data ui part schemas

• 05d2819: feat: allow zod 4.x as peer dependency

• f2b041e: Fix custom fetch in HttpChatTransport

• 2a62513: Fix error thrown when emptying messages in onError or onFinish

• 143c55b: feat (ai): export Chat callback types

• 9301f86: refactor (image-model): rename ImageModelV1 to ImageModelV2

• 904fa5e: feat (ai/core): add terminateOnError option to readUIMessageStream

• 0a87932: core (ai): change transcription model mimeType to mediaType

• 1675396: fix: avoid job executor deadlock when adding tool result

• 51f497d: feat (ai): step input message modification in prepareStep

• cee64b2: fix(otel): change back toolCall attributes of input/output back to args/result for compatibility

• f04ffe4: feat (ui): add onData callback to Chat

• bc24722: feat (ai): Add finishReason as a promise on StreamObjectResult to match StreamTextResult

• b6f9f3c: remove deprecated CoreTool* types

• 8aa9e20: feat: add speech with experimental_generateSpeech

• 4617fab: chore(embedding-models): remove remaining settings

• 8255639: ### Fix use with Google APIs + zod v4's .literal() schema

  Before zod@3.25.49, requests to Google's APIs failed due to a missing type in the provided schema. The problem has been resolved for the ai SDK by bumping our zod peer dependencies to ^3.25.49.

  pull request: https://github.com/vercel/ai/pull/6609

• f81c720: chore(ai): bundle dependencies in CLI binary

• cf9af6e: feat (ai): allow sync prepareStep

• ee38081: Add support for audio/webm to detect-media-type

• 2e4f9e4: feat (ai): improved error messages when using gateway

• 3e3b9df: fix (ai/mcp): better support for zero-argument MCP tools

• cda32ba: fix (ai): send start part in correct position in stream (streamText)

• 48a7606: feat (ai): support changing the system prompt in prepareSteps

• cb68df0: feat: add transcription and speech model support to provider registry

• db64cbe: fix (provider/openai): multi-step reasoning with tool calls

• 97c35c0: feat (ui): transient data parts

• 26695a3: feat (ui): add state for text and reasoning ui message parts

• 90ac328: fix (ui): tool part metadata support in ui messages

• 60132dd: fixed date formatting for updated mcp protocol version

• 4a1e0c8: fix(ai-cli): fix bundling and improve authentication error handling

• c6b64a7: feat (ai): allow async prepareRequest on HttpChatTransport

• fccf75c: update mcp protocol version

• 9121250: Expose provider metadata as an attribute on exported OTEL spans

• ea27cc6: chore (ai): use JSONValue definition from provider

• 90ca2b9: feat(ai): Record tool call errors on tool call spans recorded in generateText and streamText.

• 50f0362: fix (ai): fix experimental sendStart/sendFinish options in streamText

• 825e8d7: release alpha.5

• 7d97ab6: release alpha.4

• 0ff02bb: chore(provider-utils): move over jsonSchema

• 4f3776c: feat (ai): add InferUITools helper

• 9338f3e: fix (ai): throw error for v1 models

• 92c8e66: fix(ai/core): properly handle custom separator in provider registry

• 53569b8: feat (ai): add experimental repairText function to streamObject

• 82aa95d: fix (ai): merge data ui stream parts correctly

• e7d2ce3: feat: provider-executed tools

• add5ac1: feat (ai): make streamText toUIMessageStream async iterable

• 37a916d: feat (ai): add prepareSteps to streamText

• 30ac566: fix (ui): text message metadata support in ui messages

• 8026705: fix (core): send buffered text in smooth stream when stream parts change

• 9bd5ab5: feat (provider): add providerMetadata to ImageModelV2 interface (#5977)

  The experimental_generateImage method from the ai package now returnes revised prompts for OpenAI's image models.

  const prompt = "Santa Claus driving a Cadillac";
  
  const { providerMetadata } = await experimental_generateImage({
    model: openai.image("dall-e-3"),
    prompt,
  });
  
  const revisedPrompt = providerMetadata.openai.images[0]?.revisedPrompt;
  
  console.log({
    prompt,
    revisedPrompt,
  });
  

• ec5933d: chore (ai/mcp): add assertCapability method to experimental MCP client

• 09f41ac: fix (ui): add message metadata in Chat.sendMessage

• ff1c81a: feat (ai): add streamText onAbort callback

• af1d5a5: fix(ai): Unexpected reasoning-start event in extract reasoning middleware

• cb3b9c9: fix (ai): catch errors in ui message stream

• 86293e5: fix (ai): use correct generateMessageId in streamText toUIMessageStream

• d1a034f: feature: using Zod 4 for internal stuff

• fd65bc6: chore(embedding-model-v2): rename rawResponse to response

• d92b9a8: fix(ai): add support for MCP protocol version 2025-06-18

• 102b066: fix (ai): fix invalid fetch call

• 142576e: feat (ui): support message replacement in chat via messageId param on sendMessage

• 84343eb: fix (ui): call sendAutomaticallyWhen with updated messages

• a76a62b: feat (ai): add experimental prepareStep callback to generateText

• 89ba235: fix (ui): support tool names with dash

• 8e31d46: feat (ai): export SourceDocumentUIPart

• bd398e4: fix (core): improve error handling in streamText's consumeStream method

• 88a8ee5: fix (ai): support abort during retry waits

• 205077b: fix: improve Zod compatibility

• d91b50d: chore(ui-utils): merge into ai package

• e4c8647: feat (ui): allow asynchronous onFinish in createUIMessageStream

• c808e4d: fix (ui): do not send changing assistant message ids when onFinish is provided

• e862b5b: feat (ai): allow sync tool.execute

• 395c85e: feat (ai): add consumeSseStream option to UI message stream responses

• 5bdff05: Removed deprecated options.throwErrorForEmptyVectors from cosineSimilarity(). Since throwErrorForEmptyVectors was the only option the entire options argument was removed.

  - cosineSimilarity(vector1, vector2, options)
  +cosineSimilarity(vector1, vector2)
  

• 13b4f46: feat (ai): export experimental MCPClient and MCPClientConfig interfaces

• a4f3007: chore: remove ai/react

• 8e64e9c: feat (ai): allow using provider default temperature by specifying null

• b983b51: feat (ai): support model message array in prompt

• 56c232b: fix (ai): remove outdated sendStart jsdoc

• 7324c21: fix (ai/telemetry): Avoid JSON.stringify on Uint8Arrays for telemetry

• f10304b: feat(tool-calling): don't require the user to have to pass parameters

• dd5fd43: feat (ai): support dynamic tools in Chat onToolCall

• a753b3a: feat (provider/anthropic): cache control for tools

• 383cbfa: feat (ai): add isAborted to onFinish callback for ui message streams

• 27deb4d: feat (provider/gateway): Add providerMetadata to embeddings response

• 5f2b3d4: chore (ai): stable prepareStep

• 4c8f834: feat: automatically respect rate limit headers in retry logic

  Added automatic support for respecting rate limit headers (retry-after-ms and retry-after) in the SDK's retry logic. When these headers are present and contain reasonable values (0-60 seconds), the retry mechanism will use the server-specified delay instead of exponential backoff. This matches the behavior of Anthropic and OpenAI client SDKs and improves rate limit handling without requiring any API changes.

• f2c7f19: feat (ui): add Chat.clearError()

• 7bd025b: fix (ai): fix sync tool execute with streamText

• Updated dependencies

  • @ai-sdk/provider-utils@3.0.0
  • @ai-sdk/provider@2.0.0
  • @ai-sdk/gateway@1.0.0