ai 7.0.162 → 7.0.164

package/docs/03-ai-sdk-core/60-telemetry.mdx

@@ -74,6 +74,13 @@ By default, both inputs and outputs are recorded. You can disable them by settin
 Disabling the recording of inputs and outputs can be useful for privacy, data transfer, and performance reasons.
 You might for example want to disable recording inputs if they contain sensitive information.
 
+These settings filter content before lifecycle events reach telemetry integrations
+and Node.js `ai:telemetry` subscribers. Raw provider bodies and provider metadata
+are omitted when either recording option is disabled. Raw tracing span results
+are also omitted in that case because they can contain both inputs and outputs.
+Tracing span errors use a generic failure message.
+Application lifecycle callbacks and returned results retain their full content.
+
 For `generateSpeech`, input telemetry contains the source text and output
 telemetry contains the generated audio byte length, media type, and format.
 For `transcribe` and `experimental_streamTranscribe`, telemetry contains the
@@ -102,6 +109,23 @@ such as `gen_ai.usage.characters`, `gen_ai.usage.seconds`, and
 callbacks, and can also be included as the supplemental `ai.response.usage`
 attribute by enabling `usage: true`.
 
+### Provider endpoints
+
+For completed language-model calls, providers can report the known HTTP request
+endpoint. `OpenTelemetry` records its host as `server.address` and its port as
+`server.port` on the `chat` span. This includes default ports (443 for HTTPS and
+80 for HTTP) and works when input and output recording are disabled.
+
+OpenAI Chat reports this information for both `generateText` and `streamText`,
+including custom `baseURL` settings. Azure Chat and Bedrock Mantle Chat reuse
+this adapter and also report their configured endpoint. Other providers leave these attributes
+unset until they report their endpoint; no destination is inferred from the
+provider or model name, including an unknown backend behind an intermediary.
+
+The model-call end event's optional `request.url` contains only the HTTP origin.
+Credentials, paths, queries, and fragments are removed before telemetry events
+are emitted. The OpenTelemetry attributes contain only host and port.
+
 ### Opting out
 
 Telemetry is opt-out. To disable telemetry for a specific call, set `isEnabled: false`:
@@ -594,7 +618,7 @@ metrics. The `LegacyOpenTelemetry` integration does not emit this histogram.
 
 #### generateText / streamText
 
-For `generateText` and `streamText`, the integration records 3 types of spans:
+For `generateText` and `streamText`, the integration records 4 types of spans:
 
 - **`invoke_agent {modelId}`** (default root span name, `INTERNAL`): covers the full operation including all steps and tool calls. Use [`agentSpanName`](#agent-span-names) to customize the name.
 
@@ -622,7 +646,16 @@ For `generateText` and `streamText`, the integration records 3 types of spans:
   - `gen_ai.usage.cache_creation.input_tokens`: cached input tokens created
   - `gen_ai.output.messages`: the output in [GenAI SemConv message format](#genai-message-format) (when `recordOutputs` is enabled)
 
-- **`chat {modelId}`** (step span, `CLIENT`): one span per LLM provider call, nested under the root span.
+- **`step {n}`** (step span, `INTERNAL`): one span per step, nested under the root span. `n` is the step number, starting at 1. It contains the model call and tool spans of that step.
+
+  Initial attributes:
+  - `gen_ai.operation.name`: `"agent_step"`
+
+  `agent_step` is an AI SDK-defined custom operation name, not a well-known
+  OpenTelemetry GenAI operation. It marks an orchestration step, while the
+  nested `chat` span marks the model call.
+
+- **`chat {modelId}`** (model call span, `CLIENT`): one span per LLM provider call, nested under the step span.
 
   Initial attributes:
   - `gen_ai.operation.name`: `"chat"`
@@ -655,11 +688,11 @@ For `generateText` and `streamText`, the integration records 3 types of spans:
   `streamText` with the `output` property instead.
 </Note>
 
-The deprecated object APIs emit the same span hierarchy as `generateText`/`streamText` with these additional attributes on the root span:
+The deprecated object APIs emit the same spans as `generateText`/`streamText`, except that there is no `step {n}` span: the `chat {modelId}` span is nested directly under the root span. The root span has these additional attributes:
 
 - `gen_ai.output.type`: `"json"`
 
-The step spans also include `gen_ai.output.type: "json"`, and `gen_ai.output.messages` contains the generated object as a text part.
+The `chat {modelId}` spans also include `gen_ai.output.type: "json"`, and `gen_ai.output.messages` contains the generated object as a text part.
 
 #### embed / embedMany
 

package/docs/04-ai-sdk-ui/50-stream-protocol.mdx

@@ -425,6 +425,19 @@ data: {"type":"tool-output-denied","toolCallId":"call_fJdQDqnXeGxTmr4E3YPSR7Ar"}
 
 ```
 
+### Steering Message Part
+
+Experimental. A user message that was added to the run with [steering](/docs/agents/loop-control#steering). It is sent before the `start-step` part of the step that received it, and `useChat` adds it to the assistant message as a `steering-message` part.
+
+Format: Server-Sent Event with JSON object
+
+Example:
+
+```
+data: {"type":"steering-message","id":"steer-1","parts":[{"type":"text","text":"I meant Paris, Texas."}]}
+
+```
+
 ### Start Step Part
 
 A part indicating the start of a step.

package/docs/06-advanced/04-caching.mdx

@@ -14,6 +14,7 @@ and the [`simulateReadableStream`](/docs/reference/ai-sdk-core/simulate-readable
 
 Language model middleware is a way to enhance the behavior of language models by intercepting and modifying the calls to the language model.
 Let's see how you can use language model middleware to cache responses.
+Include the provider and model ID in the cache key alongside the call parameters to keep cached responses separate across models and providers.
 
 ```ts filename="ai/middleware.ts"
 import { Redis } from '@upstash/redis';
@@ -31,8 +32,12 @@ const redis = new Redis({
 
 export const cacheMiddleware: LanguageModelV4Middleware = {
   specificationVersion: 'v4',
-  wrapGenerate: async ({ doGenerate, params }) => {
-    const cacheKey = JSON.stringify(params);
+  wrapGenerate: async ({ doGenerate, params, model }) => {
+    const cacheKey = JSON.stringify({
+      provider: model.provider,
+      modelId: model.modelId,
+      params,
+    });
 
     const cached = (await redis.get(cacheKey)) as Awaited<
       ReturnType<LanguageModelV4['doGenerate']>
@@ -56,8 +61,12 @@ export const cacheMiddleware: LanguageModelV4Middleware = {
 
     return result;
   },
-  wrapStream: async ({ doStream, params }) => {
-    const cacheKey = JSON.stringify(params);
+  wrapStream: async ({ doStream, params, model }) => {
+    const cacheKey = JSON.stringify({
+      provider: model.provider,
+      modelId: model.modelId,
+      params,
+    });
 
     // Check if the result is in the cache
     const cached = await redis.get(cacheKey);

package/docs/07-reference/01-ai-sdk-core/01-generate-text.mdx

@@ -1277,6 +1277,11 @@ To see `generateText` in action, check out [these examples](#examples).
         {
           type: 'GenerateTextStepStartEvent',
           parameters: [
+            {
+              name: 'callId',
+              type: 'string',
+              description: 'Unique identifier for the generation call.',
+            },
             {
               name: 'stepNumber',
               type: 'number',
@@ -1339,40 +1344,12 @@ To see `generateText` in action, check out [these examples](#examples).
               description:
                 'Additional provider-specific options for this step.',
             },
-            {
-              name: 'timeout',
-              type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number } | undefined',
-              description:
-                'Timeout configuration for the generation. Can be a number (milliseconds) or an object with totalMs, stepMs, chunkMs.',
-            },
-            {
-              name: 'headers',
-              type: 'Record<string, string | undefined> | undefined',
-              description: 'Additional HTTP headers sent with the request.',
-            },
-            {
-              name: 'stopWhen',
-              type: 'StopCondition<TOOLS> | Array<StopCondition<TOOLS>> | undefined',
-              description:
-                'Condition(s) for stopping the generation. When the condition is an array, any of the conditions can be met to stop.',
-            },
             {
               name: 'output',
               type: 'OUTPUT | undefined',
               description:
                 'The output specification for structured outputs, if configured.',
             },
-            {
-              name: 'abortSignal',
-              type: 'AbortSignal | undefined',
-              description: 'Abort signal for cancelling the operation.',
-            },
-            {
-              name: 'include',
-              type: '{ requestBody?: boolean; requestMessages?: boolean; responseBody?: boolean } | undefined',
-              description:
-                'Settings for controlling what data is included in step results.',
-            },
             {
               name: 'runtimeContext',
               type: 'CONTEXT',
@@ -1930,6 +1907,13 @@ To see `generateText` in action, check out [these examples](#examples).
                 {
                   type: 'LanguageModelRequestMetadata',
                   parameters: [
+                    {
+                      name: 'url',
+                      type: 'string',
+                      isOptional: true,
+                      description:
+                        'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+                    },
                     {
                       name: 'messages',
                       type: 'Array<ModelMessage>',
@@ -2386,6 +2370,13 @@ To see `generateText` in action, check out [these examples](#examples).
                 {
                   type: 'LanguageModelRequestMetadata',
                   parameters: [
+                    {
+                      name: 'url',
+                      type: 'string',
+                      isOptional: true,
+                      description:
+                        'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+                    },
                     {
                       name: 'messages',
                       type: 'Array<ModelMessage>',
@@ -2811,6 +2802,13 @@ To see `generateText` in action, check out [these examples](#examples).
         {
           type: 'LanguageModelRequestMetadata',
           parameters: [
+            {
+              name: 'url',
+              type: 'string',
+              isOptional: true,
+              description:
+                'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+            },
             {
               name: 'messages',
               type: 'Array<ModelMessage>',
@@ -3192,6 +3190,13 @@ To see `generateText` in action, check out [these examples](#examples).
                 {
                   type: 'LanguageModelRequestMetadata',
                   parameters: [
+                    {
+                      name: 'url',
+                      type: 'string',
+                      isOptional: true,
+                      description:
+                        'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+                    },
                     {
                       name: 'messages',
                       type: 'Array<ModelMessage>',

package/docs/07-reference/01-ai-sdk-core/02-stream-text.mdx

@@ -1032,7 +1032,7 @@ retain the content their contracts require.
               name: 'chunk',
               type: 'TextStreamPart<TOOLS>',
               description:
-                'The stream part. This is the same union as `stream` and includes `start`, `start-step`, `text-start`, `text-delta`, `text-end`, `reasoning-start`, `reasoning-delta`, `reasoning-end`, `custom`, `source`, `file`, `reasoning-file`, `tool-call`, `tool-input-start`, `tool-input-delta`, `tool-input-end`, `tool-result`, `tool-error`, `tool-output-denied`, `tool-approval-request`, `tool-approval-response`, `finish-step`, `finish`, `abort`, `error`, and `raw`.',
+                'The stream part. This is the same union as `stream` and includes `start`, `steering-message`, `start-step`, `text-start`, `text-delta`, `text-end`, `reasoning-start`, `reasoning-delta`, `reasoning-end`, `custom`, `source`, `file`, `reasoning-file`, `tool-call`, `tool-input-start`, `tool-input-delta`, `tool-input-end`, `tool-result`, `tool-error`, `tool-output-denied`, `tool-approval-request`, `tool-approval-response`, `finish-step`, `finish`, `abort`, `error`, `raw`, and `ping`.',
               properties: [
                 {
                   type: 'TextStreamPart',
@@ -3017,6 +3017,12 @@ retain the content their contracts require.
         {
           type: 'LanguageModelRequestMetadata',
           parameters: [
+            {
+              name: 'url',
+              type: 'string',
+              isOptional: true,
+              description: 'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+            },
             {
               name: 'messages',
               type: 'Array<ModelMessage>',
@@ -3309,6 +3315,13 @@ retain the content their contracts require.
                 {
                   type: 'RequestMetadata',
                   parameters: [
+                    {
+                      name: 'url',
+                      type: 'string',
+                      isOptional: true,
+                      description:
+                        'The HTTP origin reported by the provider, without URL credentials, paths, queries or fragments.',
+                    },
                     {
                       name: 'messages',
                       type: 'Array<ModelMessage>',
@@ -3418,6 +3431,17 @@ retain the content their contracts require.
       description:
         'A stream with all events, including text deltas, tool calls, tool results, and errors. Well-formed mid-stream provider failures are emitted in error parts as StreamProviderError instances. You can use it as either an AsyncIterable or a ReadableStream. Only errors that stop the stream, such as network errors, are thrown.',
       properties: [
+        {
+          type: 'TextStreamPingPart',
+          parameters: [
+            {
+              name: 'type',
+              type: "'ping'",
+              description:
+                'A transient provider heartbeat. It contains no generated content, does not reset model-output timeouts, and is omitted from text and UI message streams. Providers may require an explicit opt-in.',
+            },
+          ],
+        },
         {
           type: 'TextStreamTextEndPart',
           description: 'End of a text block, including optional citations supporting that block.',
@@ -3677,6 +3701,33 @@ retain the content their contracts require.
             },
           ],
         },
+        {
+          type: 'TextStreamPart',
+          parameters: [
+            {
+              name: 'type',
+              type: "'steering-message'",
+              description:
+                'Experimental. A user message that was added to the run with `experimental_steering`. It is streamed before the `start-step` part of the step that received it.',
+            },
+            {
+              name: 'id',
+              type: 'string',
+              description: 'The id of the steering message.',
+            },
+            {
+              name: 'message',
+              type: 'UserModelMessage',
+              description: 'The user message that was added to the run.',
+            },
+            {
+              name: 'stepNumber',
+              type: 'number',
+              description:
+                'The number of the step whose model call received the message.',
+            },
+          ],
+        },
         {
           type: 'TextStreamPart',
           parameters: [
@@ -3694,6 +3745,12 @@ retain the content their contracts require.
                 {
                   type: 'LanguageModelRequestMetadata',
                   parameters: [
+                    {
+                      name: 'url',
+                      type: 'string',
+                      isOptional: true,
+                      description: 'HTTP origin reported by the provider, without URL credentials, paths, queries or fragments. Unset when unavailable.',
+                    },
                     {
                       name: 'messages',
                       type: 'Array<ModelMessage>',

package/docs/07-reference/01-ai-sdk-core/11-transcribe.mdx

@@ -38,7 +38,15 @@ console.log(transcript);
     {
       name: 'audio',
       type: 'DataContent (string | Uint8Array | ArrayBuffer | Buffer) | URL',
-      description: 'The audio file to generate the transcript from.',
+      description:
+        'The audio file to generate the transcript from. HTTP(S) URLs are sent directly to models with native URL support unless a custom download function is supplied; other models download the audio locally.',
+    },
+    {
+      name: 'timestampGranularities',
+      type: "Array<'word' | 'segment'>",
+      isOptional: true,
+      description:
+        'Word and segment timings to request. Supported by the OpenAI, Azure Speech, Vertex Gemini, and xAI adapters. Explicit provider options take precedence; these adapters warn about unavailable requested timings. AI Gateway warns that it does not forward this option; use provider options or a direct provider model. Omit this option to keep provider defaults. An empty array requests no additional timing detail; it does not suppress timings returned by default.',
     },
     {
       name: 'providerOptions',
@@ -76,7 +84,7 @@ console.log(transcript);
       type: '(options: { url: URL; abortSignal?: AbortSignal }) => Promise<{ data: Uint8Array; mediaType: string | undefined }>',
       isOptional: true,
       description:
-        'Custom download function for fetching audio from URLs. Use `createDownload()` from `ai` to create a download function with custom size limits, e.g. `createDownload({ maxBytes: 50 * 1024 * 1024 })`. Default: built-in download with 2 GiB limit.',
+        'Custom download function for fetching audio from URLs. Supplying this option forces local downloading even for models with native URL support. Use `createDownload()` from `ai` to create a download function with custom size limits, e.g. `createDownload({ maxBytes: 50 * 1024 * 1024 })`. Default: native URL transcription when supported, otherwise built-in download with 2 GiB limit.',
     },
   ]}
 />
@@ -96,6 +104,13 @@ console.log(transcript);
       description:
         'An array of transcript segments, each containing a portion of the transcribed text along with its start and end times in seconds.',
     },
+    {
+      name: 'words',
+      type: 'Array<{ text: string; startSecond: number; endSecond: number }>',
+      isOptional: true,
+      description:
+        'Word-level timings when the adapter exposes them, including timings returned by default. Undefined when unavailable. Existing segments behavior is preserved: some providers also put word timings in segments.',
+    },
     {
       name: 'language',
       type: 'string | undefined',

package/docs/07-reference/01-ai-sdk-core/13-generate-video.mdx

@@ -220,7 +220,8 @@ console.log(videos);
     {
       name: 'video',
       type: 'GeneratedFile',
-      description: 'The first video that was generated.',
+      description:
+        'The first video available to the SDK. For upload-only results, accessing this property throws NoVideoGeneratedError; use uploadedVideos instead.',
       properties: [
         {
           type: 'GeneratedFile',
@@ -273,6 +274,12 @@ console.log(videos);
         },
       ],
     },
+    {
+      name: 'uploadedVideos',
+      type: 'number',
+      description:
+        'Number of videos delivered directly to caller-managed storage by the provider. These are not downloaded and are not included in videos.',
+    },
     {
       name: 'warnings',
       type: 'Warning[]',

package/docs/07-reference/01-ai-sdk-core/31-ui-message.mdx

@@ -320,6 +320,29 @@ type DataUIPart<DATA_TYPES extends UIDataTypes> = ValueOf<{
 }>;
 ```
 
+### `SteeringMessageUIPart`
+
+<Note>
+  This part is experimental and exported as
+  `Experimental_SteeringMessageUIPart`.
+</Note>
+
+A user message that was added to the run with [steering](/docs/agents/loop-control#steering) while the assistant message was generated. It is placed before the `step-start` part of the step that received it.
+
+```typescript
+type SteeringMessageUIPart = {
+  type: 'steering-message';
+  /**
+   * The id of the steering message.
+   */
+  id: string;
+  /**
+   * The content of the user message.
+   */
+  parts: Array<TextUIPart | FileUIPart>;
+};
+```
+
 ### `StepStartUIPart`
 
 A step boundary part of a message.

package/docs/07-reference/02-ai-sdk-ui/05-use-realtime.mdx

@@ -13,7 +13,8 @@ Creates a browser-side realtime session for bidirectional audio and text
 conversations with a realtime provider model.
 
 The hook supports token-based provider WebSockets, application-owned WebSocket
-relays, and optional WebRTC where the model declares support. It provides controls for capture and playback, plus turn-based text input
+relays, optional WebRTC, and provider-owned transports where the model declares
+support. It provides controls for capture and playback, plus supported turn-based text input
 and tool output. Turn-based conversation messages use `UIMessage[]`; continuous
 transcript fragments remain in `session`.
 
@@ -184,12 +185,33 @@ updates remain unsupported.
 
 ## Lifecycle and failure handling
 
+### Provider-owned sessions
+
+For [Mistral voice agents](/providers/ai-sdk-providers/mistral#realtime-voice-agents),
+use `api: { provider: '/api/mistral-realtime' }`. The Mistral provider handles
+LiveKit internally.
+The hook calls `POST` on that endpoint to create a session and
+`DELETE ?sessionId=...` to close it. The server uses
+`mistral.experimental_realtime.createSession()` and `closeSession()` with its API
+key. Keep the model, voice, instructions, and allowed settings on the server.
+
+`connect()` starts microphone capture and remote audio playback. Mistral
+transcripts populate `messages`; audio travels through the provider's LiveKit
+transport. Await `close()` to finish server-side cleanup. `disconnect()` and
+unmount release local media and initiate cleanup without waiting. Client tools
+use `onToolCall` and `addToolOutput`. Text, manual audio-buffer, and response
+commands are unsupported. See the provider guide for the complete route and
+client example.
+
+### Shared lifecycle
+
 Use `status === 'connected'` as the readiness signal; `connect()` does not promise
 to wait for provider readiness. Operational startup errors are reported through
 `onError` and `status`; the legacy resolve-and-report behavior is preserved.
 
-`close()` stops local capture and submissions, then drains events until the provider
-confirms final usage or the close deadline expires. A failed close-command send
+For continuous Live sessions, `close()` stops local capture and submissions, then
+drains events until the provider confirms final usage or the close deadline
+expires. A failed close-command send
 uses the shorter accepted-event drain rather than waiting for an acknowledgment
 to an unsent command. Read `session.finalization`: a fulfilled close promise does
 not by itself confirm usage. `disconnect()` and unmount release resources immediately.
@@ -244,9 +266,9 @@ and completion of delegated work do not prove that the corresponding audio was h
     },
     {
       name: 'api',
-      type: '{ token: string } | { websocket: string; protocols?: string[] } | { session: string }',
+      type: '{ token: string } | { websocket: string; protocols?: string[] } | { session: string } | { provider: string }',
       description:
-        'Choose one supported connection mechanism: token setup, a WebSocket relay, or a WebRTC SDP endpoint. Conversation semantics and connection capabilities are independent.',
+        'Choose one supported connection mechanism: token setup, a WebSocket relay, a WebRTC SDP endpoint, or a provider-managed session endpoint. Conversation semantics and connection capabilities are independent.',
       properties: [
         {
           type: 'Object',
@@ -265,6 +287,13 @@ and completion of delegated work do not prove that the corresponding audio was h
               description:
                 'Application WebRTC setup endpoint. Accepts JSON { sdp, sessionConfig } and returns JSON { sdp, sessionId }.',
             },
+            {
+              name: 'provider',
+              type: 'string',
+              isOptional: true,
+              description:
+                'Application endpoint for a provider-owned transport. Mistral uses POST to create a session and DELETE with a sessionId query parameter to close it.',
+            },
             {
               name: 'websocket',
               type: 'string',
@@ -396,19 +425,19 @@ and completion of delegated work do not prove that the corresponding audio was h
       name: 'connect',
       type: '(options?: { stream?: MediaStream; capture?: boolean }) => Promise<void>',
       description:
-        'Starts the selected connection. Continuous browser sessions acquire microphone audio by default; capture: false opts out. A supplied Live stream remains caller-owned. The zero-argument overload remains usable as an event handler.',
+        'Starts the selected connection. Continuous Live and Mistral sessions acquire microphone audio by default; capture: false opts out. Supplied tracks remain caller-owned on these paths. The zero-argument overload remains usable as an event handler.',
     },
     {
       name: 'close',
       type: '(options?: { eventId?: string }) => Promise<void>',
       description:
-        'Stops capture/submissions and waits for bounded graceful finalization where supported. Inspect session.finalization for confirmation; models without acknowledged close disconnect immediately.',
+        'Stops capture/submissions and waits for bounded graceful finalization where supported. Provider-owned transports await session cleanup. For Live, inspect session.finalization for usage confirmation; models without acknowledged close disconnect immediately.',
     },
     {
       name: 'disconnect',
       type: '() => void',
       description:
-        'Immediately releases transports and owned media. Preserves latest usage as unconfirmed unless a terminal provider event was received.',
+        'Immediately releases transports and owned media. Provider-owned transports initiate server cleanup without awaiting it. Preserves latest usage as unconfirmed unless a terminal provider event was received.',
     },
     {
       name: 'addToolOutput',

package/docs/07-reference/04-ai-sdk-workflow/01-workflow-agent.mdx

@@ -495,7 +495,7 @@ console.log(result.text);
 console.log(result.usage);
 ```
 
-Accepts `WorkflowAgentGenerateOptions`, sharing prompt/messages, instructions, generation settings, tools and runtime context, preparation, stopping, output, and lifecycle options with `stream()`. It does not accept `writable`, `includeRawChunks`, `experimental_transform`, `sendFinish`, or `preventClose`. Without a configured stop condition, generation stops after at most 20 steps. `stream()` retains its existing unbounded default.
+Accepts `WorkflowAgentGenerateOptions`, sharing prompt/messages, instructions, generation settings, tools and runtime context, preparation, stopping, output, and lifecycle options with `stream()`. It does not accept `writable`, `includeRawChunks`, `sendFinish`, or `preventClose`. Without a configured stop condition, generation stops after at most 20 steps. `stream()` retains its existing unbounded default.
 
 The `include` option accepts `requestBody`, `responseBody`, and `requestMessages` booleans, all defaulting to `false`. Enable them only when the additional request/response data is needed. Response IDs, timestamps, model IDs, headers, and provider metadata are retained when supplied by the model.
 
@@ -626,12 +626,6 @@ const result = await agent.stream({
       isOptional: true,
       description: "Callback to attempt automatic recovery when a tool call cannot be parsed. Mark the function with 'use step' when used in a workflow.",
     },
-    {
-      name: 'experimental_transform',
-      type: 'StreamTextTransform | Array<StreamTextTransform>',
-      isOptional: true,
-      description: "Factories for stream transformations, applied in order to raw provider events before model processing and writable output. Each factory must return a TransformStream synchronously and maintain the stream structure. Mark each factory with 'use step' when used in a workflow.",
-    },
     {
       name: 'experimental_download',
       type: 'DownloadFunction',
@@ -817,48 +811,6 @@ The callback can be asynchronous. Return the repaired tool call, or `null`
 if the repair cannot be completed. See [tool call repair](/docs/ai-sdk-core/tools-and-tool-calling#tool-call-repair)
 for repair strategies.
 
-### `experimental_transform` in a workflow
-
-Pass a transform factory, or an array of factories, to `stream()`.
-Mark each factory with `'use step'` so Workflow can load it inside the model
-step. The factory must return a `TransformStream` synchronously. Keep any
-captured values serializable.
-
-```ts
-import type { LanguageModelV4StreamPart } from '@ai-sdk/provider';
-import { WorkflowAgent } from '@ai-sdk/workflow';
-import { getWritable } from 'workflow';
-
-function uppercaseText() {
-  'use step';
-
-  return new TransformStream<
-    LanguageModelV4StreamPart,
-    LanguageModelV4StreamPart
-  >({
-    transform(part, controller) {
-      controller.enqueue(
-        part.type === 'text-delta'
-          ? { ...part, delta: part.delta.toUpperCase() }
-          : part,
-      );
-    },
-  });
-}
-
-export async function uppercaseWorkflow() {
-  'use workflow';
-
-  const agent = new WorkflowAgent({ model: 'openai/gpt-5' });
-
-  return agent.stream({
-    messages: [{ role: 'user', content: 'Write a short greeting.' }],
-    writable: getWritable(),
-    experimental_transform: uppercaseText,
-  });
-}
-```
-
 ## Utilities
 
 ### `createModelCallToUIChunkTransform(options?)`