ai 7.0.142 → 7.0.143

package/dist/index.js

@@ -1174,7 +1174,7 @@ function cloneValue(value) {
 }
 //#endregion
 //#region src/version.ts
-const VERSION = "7.0.142";
+const VERSION = "7.0.143";
 //#endregion
 //#region src/util/download/download.ts
 /**
@@ -8132,6 +8132,46 @@ function invokeToolCallbacksFromStream({ stream, tools, stepInputMessages, abort
 	} }));
 }
 //#endregion
+//#region src/generate-text/stream-text-single-consumer-result.ts
+/** Expose only the stream and completion metadata for a single consumer. */
+function createSingleConsumerStreamTextResult(result) {
+	return {
+		get textStream() {
+			return result.textStream;
+		},
+		get stream() {
+			return result.stream;
+		},
+		get fullStream() {
+			return result.fullStream;
+		},
+		get partialOutputStream() {
+			return result.partialOutputStream;
+		},
+		get elementStream() {
+			return result.elementStream;
+		},
+		get totalUsage() {
+			return result.totalUsage;
+		},
+		get usage() {
+			return result.usage;
+		},
+		get finishReason() {
+			return result.finishReason;
+		},
+		get rawFinishReason() {
+			return result.rawFinishReason;
+		},
+		consumeStream: result.consumeStream.bind(result),
+		toUIMessageStream: result.toUIMessageStream.bind(result),
+		toUIMessageStreamResponse: result.toUIMessageStreamResponse.bind(result),
+		toTextStreamResponse: result.toTextStreamResponse.bind(result),
+		pipeUIMessageStreamToResponse: result.pipeUIMessageStreamToResponse.bind(result),
+		pipeTextStreamToResponse: result.pipeTextStreamToResponse.bind(result)
+	};
+}
+//#endregion
 //#region src/prompt/normalize-stream-provider-error.ts
 /**
 * Normalizes well-formed provider error payloads without changing existing
@@ -8259,7 +8299,7 @@ const originalGenerateCallId$8 = createIdGenerator$1({
 *
 * @returns A stream of model call parts together with request and response metadata when available.
 */
-async function streamLanguageModelCall({ model, tools, toolOrder, output, toolChoice, prompt, system, instructions, messages, allowSystemInMessages, download, abortSignal, headers, includeRawChunks, providerOptions, repairToolCall, refineToolInput, executeLanguageModelCallInTelemetryContext = async ({ execute }) => await execute(), callId, toolsContext, experimental_sandbox: sandbox, _internal: { generateId = originalGenerateId$3, generateCallId = originalGenerateCallId$8, now: now$3 = now } = {}, onStart, onLanguageModelCallStart, onLanguageModelCallEnd, ...callSettings }) {
+async function streamLanguageModelCall({ model, tools, toolOrder, output, toolChoice, prompt, system, instructions, messages, allowSystemInMessages, download, abortSignal, headers, includeRawChunks, providerOptions, repairToolCall, refineToolInput, executeLanguageModelCallInTelemetryContext = async ({ execute }) => await execute(), callId, toolsContext, experimental_sandbox: sandbox, _internal: { generateId = originalGenerateId$3, generateCallId = originalGenerateCallId$8, now: now$3 = now, collectContent = true } = {}, onStart, onLanguageModelCallStart, onLanguageModelCallEnd, ...callSettings }) {
 	const resolvedModel = resolveLanguageModel(model);
 	const effectiveCallId = callId ?? generateCallId();
 	const standardizedPrompt = await standardizePrompt({
@@ -8333,13 +8373,14 @@ async function streamLanguageModelCall({ model, tools, toolOrder, output, toolCh
 			generateId,
 			now: now$3,
 			callStartTimestampMs,
-			onLanguageModelCallEnd
+			onLanguageModelCallEnd,
+			collectContent
 		}))),
 		response,
 		request
 	};
 }
-function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ tools, instructions, messages, repairToolCall, refineToolInput, abortSignal, callId, provider, modelId, toolChoice, generateId, now, callStartTimestampMs, onLanguageModelCallEnd }) {
+function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ tools, instructions, messages, repairToolCall, refineToolInput, abortSignal, callId, provider, modelId, toolChoice, generateId, now, callStartTimestampMs, onLanguageModelCallEnd, collectContent }) {
 	const toolCallsByToolCallId = /* @__PURE__ */ new Map();
 	const modelCallContent = [];
 	const rawModelCallContent = [];
@@ -8368,6 +8409,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "text-start":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: textPartIndexes,
@@ -8380,6 +8422,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "text-delta":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: textPartIndexes,
@@ -8398,6 +8441,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "text-end":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: textPartIndexes,
@@ -8412,6 +8456,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "reasoning-start":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: reasoningPartIndexes,
@@ -8424,6 +8469,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "reasoning-delta":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: reasoningPartIndexes,
@@ -8442,6 +8488,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			case "reasoning-end":
 				upsertTextContentPart({
+					enabled: collectContent,
 					content: modelCallContent,
 					rawContent: rawModelCallContent,
 					partIndexes: reasoningPartIndexes,
@@ -8463,12 +8510,12 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 					}),
 					mediaType: chunk.mediaType
 				});
-				modelCallContent.push({
+				if (collectContent) modelCallContent.push({
 					type: chunk.type,
 					file,
 					...chunk.providerMetadata != null ? { providerMetadata: chunk.providerMetadata } : {}
 				});
-				rawModelCallContent.push(chunk);
+				if (collectContent) rawModelCallContent.push(chunk);
 				controller.enqueue({
 					type: chunk.type,
 					file,
@@ -8540,7 +8587,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			}
 			case "tool-call":
-				rawModelCallContent.push(chunk);
+				if (collectContent) rawModelCallContent.push(chunk);
 				try {
 					const toolCall = await parseToolCall({
 						toolCall: chunk,
@@ -8553,7 +8600,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 					});
 					toolCallsByToolCallId.set(toolCall.toolCallId, toolCall);
 					controller.enqueue(toolCall);
-					modelCallContent.push(toolCall);
+					if (collectContent) modelCallContent.push(toolCall);
 					if (toolCall.invalid) {
 						if (!toolCall.providerExecuted) controller.enqueue({
 							type: "tool-error",
@@ -8575,7 +8622,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				}
 				break;
 			case "tool-approval-request": {
-				rawModelCallContent.push(chunk);
+				if (collectContent) rawModelCallContent.push(chunk);
 				const toolCall = toolCallsByToolCallId.get(chunk.toolCallId);
 				if (toolCall == null) {
 					controller.enqueue({
@@ -8593,11 +8640,11 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 					toolCall
 				};
 				controller.enqueue(toolApprovalRequest);
-				modelCallContent.push(toolApprovalRequest);
+				if (collectContent) modelCallContent.push(toolApprovalRequest);
 				break;
 			}
 			case "tool-result": {
-				rawModelCallContent.push(chunk);
+				if (collectContent) rawModelCallContent.push(chunk);
 				const toolName = chunk.toolName;
 				const toolCall = toolCallsByToolCallId.get(chunk.toolCallId);
 				const toolResultPart = chunk.isError ? {
@@ -8622,7 +8669,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 					...toolCall?.toolMetadata != null ? { toolMetadata: toolCall.toolMetadata } : {}
 				};
 				controller.enqueue(toolResultPart);
-				modelCallContent.push(toolResultPart);
+				if (collectContent) modelCallContent.push(toolResultPart);
 				break;
 			}
 			case "tool-input-start": {
@@ -8653,8 +8700,8 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({ too
 				break;
 			default:
 				if (chunk.type === "custom" || chunk.type === "source") {
-					modelCallContent.push(chunk);
-					rawModelCallContent.push(chunk);
+					if (collectContent) modelCallContent.push(chunk);
+					if (collectContent) rawModelCallContent.push(chunk);
 				}
 				controller.enqueue(chunk);
 		}
@@ -8686,7 +8733,8 @@ function calculateNearestRankPercentile(sortedValues, percentile) {
 /**
 * Appends a text or reasoning content part into the content array and updates the part indexes.
 */
-function upsertTextContentPart({ content, rawContent, partIndexes, rawPartIndexes, id, type, textDelta, providerMetadata }) {
+function upsertTextContentPart({ enabled, content, rawContent, partIndexes, rawPartIndexes, id, type, textDelta, providerMetadata }) {
+	if (!enabled) return;
 	let partIndex = partIndexes.get(id);
 	if (partIndex == null) {
 		partIndex = content.push({
@@ -8726,75 +8774,26 @@ const originalGenerateCallId$7 = createIdGenerator$1({
 	prefix: "call",
 	size: 24
 });
-/**
-* Generate a text and call tools for a given prompt using a language model.
-*
-* This function streams the output. If you do not want to stream the output, use `generateText` instead.
-*
-* @param model - The language model to use.
-* @param tools - Tools that are accessible to and can be called by the model. The model needs to support calling tools.
-* @param toolOrder - Controls the order in which tools are sent to the provider. Tools not listed are appended alphabetically.
-* @param toolCallConcurrency - Maximum number of tool calls that may execute concurrently within a step. Set to 1 for sequential execution. Default: unlimited.
-*
-* @param system - A system message that will be part of the prompt.
-* @param prompt - A simple text prompt. You can either use `prompt` or `messages` but not both.
-* @param messages - A list of messages. You can either use `prompt` or `messages` but not both.
-* @param allowSystemInMessages - Whether system messages are allowed in the `prompt` or `messages` fields. Default: false.
-*
-* @param maxOutputTokens - Maximum number of tokens to generate.
-* @param temperature - Temperature setting.
-* The value is passed through to the provider. The range depends on the provider and model.
-* It is recommended to set either `temperature` or `topP`, but not both.
-* @param topP - Nucleus sampling.
-* The value is passed through to the provider. The range depends on the provider and model.
-* It is recommended to set either `temperature` or `topP`, but not both.
-* @param topK - Only sample from the top K options for each subsequent token.
-* Used to remove "long tail" low probability responses.
-* Recommended for advanced use cases only. You usually only need to use temperature.
-* @param presencePenalty - Presence penalty setting.
-* It affects the likelihood of the model to repeat information that is already in the prompt.
-* The value is passed through to the provider. The range depends on the provider and model.
-* @param frequencyPenalty - Frequency penalty setting.
-* It affects the likelihood of the model to repeatedly use the same words or phrases.
-* The value is passed through to the provider. The range depends on the provider and model.
-* @param stopSequences - Stop sequences.
-* If set, the model will stop generating text when one of the stop sequences is generated.
-* @param seed - The seed (integer) to use for random sampling.
-* If set and supported by the model, calls will generate deterministic results.
-*
-* @param maxRetries - Maximum number of retries. Set to 0 to disable retries. Default: 2.
-* @param streamRetries - Maximum number of retries for provider errors received after streaming starts. Set to 0 to disable automatic stream retries while allowing `onError` to request retries. Omit to disable all stream retry behavior. Default: 0.
-* @param abortSignal - An optional abort signal that can be used to cancel the call.
-* @param timeout - An optional timeout in milliseconds. The call will be aborted if it takes longer than the specified timeout.
-* @param headers - Additional HTTP headers to be sent with the request. Only applicable for HTTP-based providers.
-*
-* @param experimental_sandbox - The sandbox environment that is passed through to tool execution.
-* @param runtimeContext - User-defined runtime context that flows through the entire generation lifecycle.
-* @param experimental_refineToolInput - Optional mapping of tool names to functions that refine parsed tool inputs before tools are executed and before outputs, callbacks, and telemetry are recorded.
-*
-* @param onChunk - Callback that is called for each chunk of the stream. The stream processing will pause until the callback promise is resolved.
-* @param onError - Callback that is called when an error occurs during streaming. You can use it to log errors.
-* @param onStart - Callback invoked when generation begins, before any LLM calls.
-* @param experimental_onStart - Deprecated alias for `onStart`.
-* @param onStepStart - Callback invoked when each step begins, before the provider is called.
-* @param experimental_onStepStart - Deprecated alias for `onStepStart`.
-* @param onLanguageModelCallStart - Callback invoked immediately before each provider model call begins.
-* @param experimental_onLanguageModelCallStart - Deprecated alias for `onLanguageModelCallStart`.
-* @param onLanguageModelCallEnd - Callback invoked after each provider model call response is normalized and parsed.
-* @param experimental_onLanguageModelCallEnd - Deprecated alias for `onLanguageModelCallEnd`.
-* @param onToolExecutionStart - Callback invoked before each tool execution begins.
-* @param experimental_onToolCallStart - Deprecated alias for `onToolExecutionStart`.
-* @param onToolExecutionEnd - Callback invoked after each tool execution completes.
-* @param experimental_onToolCallFinish - Deprecated alias for `onToolExecutionEnd`.
-* @param onStepEnd - Callback that is called when each step (LLM call) ends, including intermediate steps.
-* @param onStepFinish - Deprecated alias for `onStepEnd`.
-* @param onEnd - Callback that is called when all steps are finished and the response is complete.
-* @param onFinish - Deprecated alias for `onEnd`.
-*
-* @returns
-* A result object for accessing different stream types and additional information.
-*/
-function streamText({ model, tools, toolChoice, instructions, system, prompt, messages, allowSystemInMessages, maxRetries, streamRetries, abortSignal, timeout, headers, stopWhen = isStepCount(1), experimental_sandbox: sandbox, output, toolApproval, experimental_toolCallers, experimental_toolApprovalSecret, experimental_telemetry, telemetry = experimental_telemetry, prepareStep, providerOptions, activeTools, toolOrder, toolCallConcurrency, experimental_repairToolCall, repairToolCall = experimental_repairToolCall, experimental_refineToolInput: refineToolInput, experimental_transform: transform, experimental_download: download, includeRawChunks, onChunk, onError: onErrorArg, onFinish, onEnd = onFinish, onAbort, onStepEnd, onStepFinish, onStart, experimental_onStart, onStepStart, experimental_onStepStart, onLanguageModelCallStart, experimental_onLanguageModelCallStart, onLanguageModelCallEnd, experimental_onLanguageModelCallEnd, onToolExecutionStart, onToolExecutionEnd, experimental_onToolCallStart, experimental_onToolCallFinish, runtimeContext = {}, toolsContext = {}, experimental_include, include = experimental_include, _internal: { now: now$2 = now, generateId = originalGenerateId$2, generateCallId = originalGenerateCallId$7 } = {}, ...settings }) {
+function streamText({ experimental_lowMemory = false, ...options }) {
+	if (!experimental_lowMemory) return streamTextInternal({ ...options });
+	const telemetry = options.telemetry ?? options.experimental_telemetry;
+	const integrations = telemetry?.integrations != null ? asArray(telemetry.integrations) : getGlobalTelemetryIntegrations();
+	const collectContent = options.tools != null && Object.keys(options.tools).length > 0 || options.toolChoice != null || options.output != null || options.prepareStep != null || options.stopWhen != null || options.onEnd != null || options.onFinish != null || options.onStepEnd != null || options.onStepFinish != null || options.onAbort != null || options.onLanguageModelCallEnd != null || options.experimental_onLanguageModelCallEnd != null || telemetry?.isEnabled !== false && integrations.length > 0;
+	return createSingleConsumerStreamTextResult(streamTextInternal({
+		...options,
+		_internal: {
+			...options._internal,
+			retention: {
+				replay: false,
+				collectContent
+			}
+		}
+	}));
+}
+function streamTextInternal({ model, tools, toolChoice, instructions, system, prompt, messages, allowSystemInMessages, maxRetries, streamRetries, abortSignal, timeout, headers, stopWhen = isStepCount(1), experimental_sandbox: sandbox, output, toolApproval, experimental_toolCallers, experimental_toolApprovalSecret, experimental_telemetry, telemetry = experimental_telemetry, prepareStep, providerOptions, activeTools, toolOrder, toolCallConcurrency, experimental_repairToolCall, repairToolCall = experimental_repairToolCall, experimental_refineToolInput: refineToolInput, experimental_transform: transform, experimental_download: download, includeRawChunks, onChunk, onError: onErrorArg, onFinish, onEnd = onFinish, onAbort, onStepEnd, onStepFinish, onStart, experimental_onStart, onStepStart, experimental_onStepStart, onLanguageModelCallStart, experimental_onLanguageModelCallStart, onLanguageModelCallEnd, experimental_onLanguageModelCallEnd, onToolExecutionStart, onToolExecutionEnd, experimental_onToolCallStart, experimental_onToolCallFinish, runtimeContext = {}, toolsContext = {}, experimental_include, include = experimental_include, _internal: { now: now$2 = now, generateId = originalGenerateId$2, generateCallId = originalGenerateCallId$7, retention = {
+	replay: true,
+	collectContent: true
+} } = {}, ...settings }) {
 	const resolvedToolCallConcurrency = prepareToolCallConcurrency(toolCallConcurrency);
 	const totalTimeoutMs = getTotalTimeoutMs(timeout);
 	const stepTimeoutMs = getStepTimeoutMs(timeout);
@@ -8814,6 +8813,7 @@ function streamText({ model, tools, toolChoice, instructions, system, prompt, me
 	const resolvedOnToolExecutionEnd = onToolExecutionEnd ?? experimental_onToolCallFinish;
 	const resolvedOnStepEnd = onStepEnd ?? onStepFinish;
 	return new DefaultStreamTextResult({
+		retention,
 		model: resolveLanguageModel(model),
 		telemetry,
 		headers,
@@ -8883,7 +8883,7 @@ async function markPromiseAsHandled$1(promise) {
 		await promise;
 	} catch {}
 }
-function createOutputTransformStream(output) {
+function createOutputTransformStream(output, collectOutput = true) {
 	let firstTextChunkId = void 0;
 	let text = "";
 	let textChunk = "";
@@ -8917,6 +8917,13 @@ function createOutputTransformStream(output) {
 		textChunk = "";
 	}
 	return new TransformStream({ async transform(chunk, controller) {
+		if (!collectOutput) {
+			controller.enqueue(isStreamRetryBoundaryPart(chunk) ? chunk : {
+				part: chunk,
+				partialOutput: void 0
+			});
+			return;
+		}
 		if (isStreamRetryBoundaryPart(chunk)) {
 			resetOutputState();
 			controller.enqueue(chunk);
@@ -9047,12 +9054,14 @@ function applyStreamTextTransforms({ stream, transforms, tools, stopStream }) {
 	});
 }
 var DefaultStreamTextResult = class {
-	constructor({ model, telemetry, headers, settings, maxRetries: maxRetriesArg, streamRetries: streamRetriesArg, abortSignal, stepTimeoutMs, stepAbortController, firstChunkTimeoutMs, firstChunkAbortController, chunkTimeoutMs, chunkAbortController, instructions, system, prompt, messages, allowSystemInMessages, experimental_sandbox: sandbox, tools, toolChoice, transforms, activeTools, toolOrder, toolCallConcurrency, repairToolCall, refineToolInput, stopConditions, output, toolApproval, experimental_toolCallers, experimental_toolApprovalSecret, providerOptions, prepareStep, now, generateId, generateCallId, timeout, onChunk, onError, canRetryStreamViaOnError, onEnd, onAbort, onStepFinish, onStart, onStepStart, onLanguageModelCallStart, onLanguageModelCallEnd, onToolExecutionStart, onToolExecutionEnd, runtimeContext, toolsContext, download, include }) {
+	constructor({ retention, model, telemetry, headers, settings, maxRetries: maxRetriesArg, streamRetries: streamRetriesArg, abortSignal, stepTimeoutMs, stepAbortController, firstChunkTimeoutMs, firstChunkAbortController, chunkTimeoutMs, chunkAbortController, instructions, system, prompt, messages, allowSystemInMessages, experimental_sandbox: sandbox, tools, toolChoice, transforms, activeTools, toolOrder, toolCallConcurrency, repairToolCall, refineToolInput, stopConditions, output, toolApproval, experimental_toolCallers, experimental_toolApprovalSecret, providerOptions, prepareStep, now, generateId, generateCallId, timeout, onChunk, onError, canRetryStreamViaOnError, onEnd, onAbort, onStepFinish, onStart, onStepStart, onLanguageModelCallStart, onLanguageModelCallEnd, onToolExecutionStart, onToolExecutionEnd, runtimeContext, toolsContext, download, include }) {
 		this._totalUsage = new DelayedPromise();
 		this._finishReason = new DelayedPromise();
 		this._rawFinishReason = new DelayedPromise();
 		this._steps = new DelayedPromise();
 		this._initialResponseMessages = new DelayedPromise();
+		this.streamClaimed = false;
+		this.retention = retention;
 		this.outputSpecification = output;
 		this.tools = tools;
 		const resolvedToolCallers = resolveToolCallerConfiguration({
@@ -9076,7 +9085,7 @@ var DefaultStreamTextResult = class {
 		let recordedRequest = {};
 		let recordedRequestMessages = [];
 		let recordedWarnings = [];
-		const recordedSteps = [];
+		let recordedSteps = [];
 		const initialResponseMessages = [];
 		let stepMessagesForNextStep;
 		let currentStepMessages = [];
@@ -9134,14 +9143,16 @@ var DefaultStreamTextResult = class {
 						}
 					});
 				}
-				if (part.type === "custom" || part.type === "source" || part.type === "tool-call" || part.type === "tool-approval-request" || part.type === "tool-approval-response" || part.type === "tool-error") recordedContent.push(part);
+				if (part.type === "custom" || part.type === "source" || part.type === "tool-call" || part.type === "tool-approval-request" || part.type === "tool-approval-response" || part.type === "tool-error") {
+					if (retention.collectContent) recordedContent.push(part);
+				}
 				if (part.type === "text-start") {
 					activeTextContent[part.id] = {
 						type: "text",
 						text: "",
 						providerMetadata: part.providerMetadata
 					};
-					recordedContent.push(activeTextContent[part.id]);
+					if (retention.collectContent) recordedContent.push(activeTextContent[part.id]);
 				}
 				if (part.type === "text-delta") {
 					const activeText = activeTextContent[part.id];
@@ -9155,7 +9166,7 @@ var DefaultStreamTextResult = class {
 						});
 						return;
 					}
-					activeText.text += part.text;
+					if (retention.collectContent) activeText.text += part.text;
 					activeText.providerMetadata = part.providerMetadata ?? activeText.providerMetadata;
 				}
 				if (part.type === "text-end") {
@@ -9180,7 +9191,7 @@ var DefaultStreamTextResult = class {
 						text: "",
 						providerMetadata: part.providerMetadata
 					};
-					recordedContent.push(activeReasoningContent[part.id]);
+					if (retention.collectContent) recordedContent.push(activeReasoningContent[part.id]);
 				}
 				if (part.type === "reasoning-delta") {
 					const activeReasoning = activeReasoningContent[part.id];
@@ -9194,7 +9205,7 @@ var DefaultStreamTextResult = class {
 						});
 						return;
 					}
-					activeReasoning.text += part.text;
+					if (retention.collectContent) activeReasoning.text += part.text;
 					activeReasoning.providerMetadata = part.providerMetadata ?? activeReasoning.providerMetadata;
 				}
 				if (part.type === "reasoning-end") {
@@ -9212,12 +9223,16 @@ var DefaultStreamTextResult = class {
 					activeReasoning.providerMetadata = part.providerMetadata ?? activeReasoning.providerMetadata;
 					delete activeReasoningContent[part.id];
 				}
-				if (part.type === "file" || part.type === "reasoning-file") recordedContent.push({
-					type: part.type,
-					file: part.file,
-					...part.providerMetadata != null ? { providerMetadata: part.providerMetadata } : {}
-				});
-				if (part.type === "tool-result" && !part.preliminary) recordedContent.push(part);
+				if (part.type === "file" || part.type === "reasoning-file") {
+					if (retention.collectContent) recordedContent.push({
+						type: part.type,
+						file: part.file,
+						...part.providerMetadata != null ? { providerMetadata: part.providerMetadata } : {}
+					});
+				}
+				if (part.type === "tool-result" && !part.preliminary) {
+					if (retention.collectContent) recordedContent.push(part);
+				}
 				if (part.type === "start-step") {
 					recordedContent = [];
 					activeReasoningContent = createIdMap();
@@ -9284,7 +9299,7 @@ var DefaultStreamTextResult = class {
 					self._finishReason.resolve(finishReason);
 					self._rawFinishReason.resolve(recordedRawFinishReason);
 					self._totalUsage.resolve(totalUsage);
-					self._steps.resolve(recordedSteps);
+					self._steps.resolve(retention.replay ? recordedSteps : []);
 					if (isAborted) return;
 					const finalStep = recordedSteps[recordedSteps.length - 1];
 					const content = recordedSteps.flatMap((step) => step.content);
@@ -9298,7 +9313,7 @@ var DefaultStreamTextResult = class {
 					const dynamicToolResults = recordedSteps.flatMap((step) => step.dynamicToolResults);
 					const warnings = recordedSteps.flatMap((step) => step.warnings ?? []);
 					const onEndWithOutput = onEnd == null ? void 0 : async (event) => {
-						const parsedOutput = output == null ? void 0 : await self.getOutputPromise().catch(() => void 0);
+						const parsedOutput = output == null ? void 0 : await (retention.replay ? self.getOutputPromise() : self.parseOutput(finalStep)).catch(() => void 0);
 						await onEnd({
 							...event,
 							...output != null ? { output: parsedOutput } : {}
@@ -9343,6 +9358,16 @@ var DefaultStreamTextResult = class {
 					})]);
 				} catch (error) {
 					controller.error(error);
+				} finally {
+					if (!retention.replay) {
+						recordedSteps = [];
+						recordedContent = [];
+						recordedRequestMessages = [];
+						currentStepMessages = [];
+						stepMessagesForNextStep = void 0;
+						activeTextContent = createIdMap();
+						activeReasoningContent = createIdMap();
+					}
 				}
 			}
 		});
@@ -9421,7 +9446,7 @@ var DefaultStreamTextResult = class {
 				isRunning = false;
 			}
 		});
-		this.baseStream = stream.pipeThrough(createOutputTransformStream(output ?? text())).pipeThrough(eventProcessor);
+		this.baseStream = stream.pipeThrough(createOutputTransformStream(output ?? text(), retention.replay || output != null)).pipeThrough(eventProcessor);
 		const { maxRetries } = prepareRetries({
 			maxRetries: maxRetriesArg,
 			abortSignal
@@ -9765,7 +9790,10 @@ var DefaultStreamTextResult = class {
 								callbacks: [onStepStart, telemetryDispatcher.onStepStart]
 							});
 						},
-						_internal: { now },
+						_internal: {
+							now,
+							collectContent: retention.collectContent
+						},
 						...stepCallSettings
 					})));
 					const initialLanguageModelCall = await callLanguageModel();
@@ -10197,7 +10225,7 @@ var DefaultStreamTextResult = class {
 		});
 	}
 	get steps() {
-		this.consumeStream();
+		if (this.retention.replay) this.consumeStream();
 		return this._steps.promise;
 	}
 	get finalStep() {
@@ -10258,15 +10286,15 @@ var DefaultStreamTextResult = class {
 		return Promise.all([this._initialResponseMessages.promise, this.steps]).then(([initialResponseMessages, steps]) => [...initialResponseMessages, ...steps.flatMap((step) => step.response.messages)]);
 	}
 	get totalUsage() {
-		this.consumeStream();
+		if (this.retention.replay) this.consumeStream();
 		return this._totalUsage.promise;
 	}
 	get finishReason() {
-		this.consumeStream();
+		if (this.retention.replay) this.consumeStream();
 		return this._finishReason.promise;
 	}
 	get rawFinishReason() {
-		this.consumeStream();
+		if (this.retention.replay) this.consumeStream();
 		return this._rawFinishReason.promise;
 	}
 	/**
@@ -10278,15 +10306,50 @@ var DefaultStreamTextResult = class {
 	* However, the LLM results are expected to be small enough to not cause issues.
 	*/
 	teeStream() {
+		if (!this.retention.replay) {
+			if (this.streamClaimed) throw new AISDKError$1({
+				name: "AI_StreamAlreadyConsumedError",
+				message: "This result supports one stream consumer. Choose one stream projection or response helper."
+			});
+			this.streamClaimed = true;
+			const reader = this.baseStream.getReader();
+			const rejectResults = (error) => this.rejectResultPromises(error);
+			return new ReadableStream({
+				async pull(controller) {
+					try {
+						const { done, value } = await reader.read();
+						if (done) {
+							reader.releaseLock();
+							controller.close();
+						} else controller.enqueue(value);
+					} catch (error) {
+						rejectResults(error);
+						reader.releaseLock();
+						controller.error(error);
+					}
+				},
+				async cancel(reason) {
+					rejectResults(reason ?? new DOMException("The stream was cancelled.", "AbortError"));
+					try {
+						await reader.cancel(reason);
+					} finally {
+						reader.releaseLock();
+					}
+				}
+			}, { highWaterMark: 0 });
+		}
 		const [stream1, stream2] = this.baseStream.tee();
 		this.baseStream = stream2;
 		return stream1;
 	}
 	get textStream() {
-		return createAsyncIterableStream(toTextStream({ stream: this.stream }));
+		return this.asIterableStream(toTextStream({ stream: this.stream }));
+	}
+	asIterableStream(stream) {
+		return this.retention.replay ? createAsyncIterableStream(stream) : asAsyncIterableStream(stream);
 	}
 	get stream() {
-		return createAsyncIterableStream(this.teeStream().pipeThrough(new TransformStream({ transform({ part }, controller) {
+		return this.asIterableStream(this.teeStream().pipeThrough(new TransformStream({ transform({ part }, controller) {
 			controller.enqueue(part);
 		} })));
 	}
@@ -10322,9 +10385,10 @@ var DefaultStreamTextResult = class {
 		}
 	}
 	async consumeStream(options) {
+		const stream = this.stream;
 		try {
 			await consumeStream({
-				stream: this.stream,
+				stream,
 				onError: (error) => {
 					this.rejectResultPromises(error);
 					options?.onError?.(error);
@@ -10339,25 +10403,27 @@ var DefaultStreamTextResult = class {
 		return this.partialOutputStream;
 	}
 	get partialOutputStream() {
-		return createAsyncIterableStream(this.teeStream().pipeThrough(new TransformStream({ transform({ partialOutput }, controller) {
+		if (!this.retention.replay && this.outputSpecification == null) throw new UnsupportedFunctionalityError$1({ functionality: "partialOutputStream without an explicit output specification with experimental_lowMemory enabled" });
+		return this.asIterableStream(this.teeStream().pipeThrough(new TransformStream({ transform({ partialOutput }, controller) {
 			if (partialOutput !== void 0) controller.enqueue(partialOutput);
 		} })));
 	}
 	get elementStream() {
 		const transform = this.outputSpecification?.createElementStreamTransform();
 		if (transform == null) throw new UnsupportedFunctionalityError$1({ functionality: `element streams in ${this.outputSpecification?.name ?? "text"} mode` });
-		return createAsyncIterableStream(this.teeStream().pipeThrough(transform));
+		return this.asIterableStream(this.teeStream().pipeThrough(transform));
 	}
 	getOutputPromise() {
-		if (this.outputPromise == null) this.outputPromise = this.finalStep.then((step) => {
-			return (this.outputSpecification ?? text()).parseCompleteOutput({ text: step.text }, {
-				response: step.response,
-				usage: step.usage,
-				finishReason: step.finishReason
-			});
-		});
+		if (this.outputPromise == null) this.outputPromise = this.finalStep.then((step) => this.parseOutput(step));
 		return this.outputPromise;
 	}
+	parseOutput(step) {
+		return (this.outputSpecification ?? text()).parseCompleteOutput({ text: step.text }, {
+			response: step.response,
+			usage: step.usage,
+			finishReason: step.finishReason
+		});
+	}
 	get output() {
 		return this.getOutputPromise();
 	}
@@ -10574,7 +10640,8 @@ var ToolLoopAgent = class {
 		return await stream({
 			...preparedCall,
 			...callbackArgs,
-			headers: this.agentHeaders(preparedCall)
+			headers: this.agentHeaders(preparedCall),
+			experimental_lowMemory: false
 		});
 	}
 };
@@ -12937,6 +13004,7 @@ function validateDecisionInput({ state, questions }) {
 	for (const [id, question] of Object.entries(questions)) {
 		const parameter = `questions.${id}`;
 		if (!isRecord$1(question) || !isInput(question.instructions)) invalidInput(parameter, question, "instructions must be a JSON-compatible string, object, or array");
+		if (question.type === "box") continue;
 		const criteria = question.criteria;
 		switch (question.type) {
 			case "choice":
@@ -12949,7 +13017,7 @@ function validateDecisionInput({ state, questions }) {
 				if (criteria === void 0) continue;
 				if (!isRecord$1(criteria) || Object.keys(criteria).some((key) => key !== "true" && key !== "false")) invalidInput(parameter, question, "boolean criteria may only describe true and false");
 				break;
-			default: invalidInput(parameter, question, "question type must be choice, score, or boolean");
+			default: invalidInput(parameter, question, "question type must be choice, score, boolean, or box");
 		}
 		if (!isJSON(criteria) || Object.values(criteria).some((value) => value !== null && !isInput(value))) invalidInput(parameter, question, "criteria descriptions must be JSON-compatible strings, objects, arrays, or null");
 	}
@@ -13003,6 +13071,16 @@ function validateDecisionAnswers({ questions, answers, rounding }) {
 					if (Math.abs(mean - answer.score) > tolerance + meanRoundingError + scoreError) invalidAnswer(answers, `Question "${id}" score must equal the probability-weighted mean within the declared rounding precision.`);
 				}
 				break;
+			case "box": {
+				const box = answer.box;
+				if (!isRecord$1(box) || !hasExactKeys(box, [
+					"top",
+					"bottom",
+					"left",
+					"right"
+				]) || !isProbability(box.left) || !isProbability(box.right) || !isProbability(box.top) || !isProbability(box.bottom) || !(box.left < box.right) || !(box.top < box.bottom)) invalidAnswer(answers, `Question "${id}" must return box edges in [0, 1] with left < right and top < bottom.`);
+				break;
+			}
 			case "boolean": if (!isProbability(answer.probability)) invalidAnswer(answers, `Question "${id}" must return P(true) as a finite probability in [0, 1].`);
 		}
 	}
@@ -19849,7 +19927,10 @@ var DefaultChatTransport = class extends HttpChatTransport {
 //#endregion
 //#region src/ui/chat.ts
 var AbstractChat = class {
-	constructor({ generateId = generateId$1, id = generateId(), transport = new DefaultChatTransport(), messageMetadataSchema, dataPartSchemas, state, onError, onToolCall, onFinish, onData, sendAutomaticallyWhen }) {
+	get id() {
+		return this._id ??= this.generateId();
+	}
+	constructor({ generateId = generateId$1, id, transport = new DefaultChatTransport(), messageMetadataSchema, dataPartSchemas, state, onError, onToolCall, onFinish, onData, sendAutomaticallyWhen }) {
 		this.pendingMessagePreparations = /* @__PURE__ */ new Set();
 		this.activeResponse = void 0;
 		this.activeResumeRequest = void 0;
@@ -20056,7 +20137,7 @@ var AbstractChat = class {
 			await this.stop();
 			this.transport.close?.();
 		};
-		this.id = id;
+		this._id = id;
 		this.transport = transport;
 		this.generateId = generateId;
 		this.messageMetadataSchema = messageMetadataSchema;