@openclaw/ai 2026.7.2-beta.2 → 2026.7.2-beta.3

package/dist/{anthropic-BIRSg5x9.mjs → anthropic-D7SyZQ5v.mjs}

@@ -2,12 +2,12 @@ import { n as getEnvApiKey, r as __exportAll } from "./env-api-keys-DnhYpc27.mjs
 import { c as resolveClaudeNativeThinkingLevelMap, d as supportsClaudeNativeMaxEffort, f as supportsClaudeNativeXhighEffort, l as resolveClaudeSonnet5ModelIdentity, u as supportsClaudeAdaptiveThinking } from "./src-CXno1H5g.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-4t713IeR.mjs";
-import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Q1LSRIdo.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-1GiZ2_rr.mjs";
 import { r as createDeferredEventBuffer, t as notifyLlmRequestActivity } from "./llm-request-activity-CehVkZP-.mjs";
 import { t as headersToRecord } from "./headers-B_e4-1J0.mjs";
 import { n as parseStreamingJson, t as parseJsonWithRepair } from "./json-parse-BvXNt1-7.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { C as stripSystemPromptCacheBoundary, S as splitSystemPromptCacheBoundary, _ as usesClaudeFable5MessagesContract, c as adjustMaxTokensForThinking, d as isRecord, f as applyClaudeRequestContract, h as requiresClaudeAdaptiveThinking, i as extractToolResultText, l as buildBaseOptions, m as prepareClaudeSonnet5RequestContext, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, r as extractToolResultBlockText, t as transformMessages, v as usesClaudeStreamingRefusalContract } from "./transform-messages-BmS70CP5.mjs";
+import { C as splitSystemPromptCacheBoundary, c as adjustMaxTokensForThinking, f as isRecord, g as requiresClaudeAdaptiveThinking, h as prepareClaudeSonnet5RequestContext, i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, p as applyClaudeRequestContract, r as extractToolResultBlockText, t as transformMessages, u as clampMaxTokensToModel, v as usesClaudeFable5MessagesContract, w as stripSystemPromptCacheBoundary, y as usesClaudeStreamingRefusalContract } from "./transform-messages-eb6lxDLB.mjs";
 import { t as projectRuntimeToolInputSchema } from "./tool-schema-json-projection-BwNu3nDi.mjs";
 import { a as resolveCacheRetention, i as resolveCloudflareBaseUrl, n as hasCopilotVisionInput, t as buildCopilotDynamicHeaders } from "./github-copilot-headers-BsH5cqGj.mjs";
 import Anthropic from "@anthropic-ai/sdk";
@@ -295,6 +295,16 @@ function resolveOriginalAnthropicToolName(name, projection) {
 function readAnthropicUsageTokenCount(value) {
 	return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : void 0;
 }
+function readAnthropicCacheWriteUsage(usage) {
+	if (!usage.cache_creation || typeof usage.cache_creation !== "object") return {};
+	const cacheCreation = usage.cache_creation;
+	const cacheWrite5m = readAnthropicUsageTokenCount(cacheCreation.ephemeral_5m_input_tokens);
+	const cacheWrite1h = readAnthropicUsageTokenCount(cacheCreation.ephemeral_1h_input_tokens);
+	return {
+		...cacheWrite5m !== void 0 ? { cacheWrite5m } : {},
+		...cacheWrite1h !== void 0 ? { cacheWrite1h } : {}
+	};
+}
 function readAnthropicPromptUsageSnapshot(usage) {
 	const input = readAnthropicUsageTokenCount(usage.input_tokens);
 	const cacheRead = usage.cache_read_input_tokens == null ? 0 : readAnthropicUsageTokenCount(usage.cache_read_input_tokens);
@@ -417,7 +427,8 @@ function getAnthropicCompat(model) {
 		supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,
 		supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,
 		sendSessionAffinityHeaders: model.compat?.sendSessionAffinityHeaders ?? (isFireworks || isCloudflareAiGatewayAnthropic),
-		supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks
+		supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,
+		allowEmptySignature: model.compat?.allowEmptySignature ?? false
 	};
 }
 function mergeHeaders(...headerSources) {
@@ -516,7 +527,7 @@ const streamAnthropic = (model, context, options) => {
 			const sdkRequestOptions = {
 				...requestOptions?.signal ? { signal: requestOptions.signal } : {},
 				...requestOptions?.timeoutMs !== void 0 ? { timeout: requestOptions.timeoutMs } : {},
-				...requestOptions?.maxRetries !== void 0 ? { maxRetries: requestOptions.maxRetries } : {}
+				maxRetries: requestOptions?.maxRetries ?? 0
 			};
 			const response = await client.messages.create({
 				...params,
@@ -542,6 +553,8 @@ const streamAnthropic = (model, context, options) => {
 				if (cacheReadTokens !== void 0) output.usage.cacheRead = cacheReadTokens;
 				const cacheWriteTokens = event.message.usage.cache_creation_input_tokens == null ? 0 : readAnthropicUsageTokenCount(event.message.usage.cache_creation_input_tokens);
 				if (cacheWriteTokens !== void 0) output.usage.cacheWrite = cacheWriteTokens;
+				const cacheWriteUsage = readAnthropicCacheWriteUsage(event.message.usage);
+				if (cacheWriteUsage.cacheWrite1h !== void 0) output.usage.cacheWrite1h = cacheWriteUsage.cacheWrite1h;
 				output.usage.totalTokens = output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
 				if (messageStartPromptUsage && outputTokens !== void 0) output.usage.contextUsage = {
 					state: "available",
@@ -730,30 +743,34 @@ const streamAnthropic = (model, context, options) => {
 			} else if (event.type === "message_delta") {
 				if (event.delta.stop_reason) if (event.delta.stop_reason === "refusal") applyAnthropicRefusal(output, event.delta.stop_details, model.provider);
 				else output.stopReason = mapStopReason(event.delta.stop_reason);
-				const inputTokens = readAnthropicUsageTokenCount(event.usage.input_tokens);
-				if (inputTokens !== void 0) output.usage.input = inputTokens;
-				const outputTokens = readAnthropicUsageTokenCount(event.usage.output_tokens);
-				if (outputTokens !== void 0) output.usage.output = outputTokens;
-				const cacheReadTokens = readAnthropicUsageTokenCount(event.usage.cache_read_input_tokens);
-				if (cacheReadTokens !== void 0) output.usage.cacheRead = cacheReadTokens;
-				const cacheWriteTokens = readAnthropicUsageTokenCount(event.usage.cache_creation_input_tokens);
-				if (cacheWriteTokens !== void 0) output.usage.cacheWrite = cacheWriteTokens;
-				output.usage.totalTokens = output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
-				const iterationUsage = readLastAnthropicIterationUsage(event.usage);
-				if (iterationUsage.state === "valid") output.usage.contextUsage = {
-					state: "available",
-					promptTokens: iterationUsage.usage.contextPromptTokens,
-					totalTokens: iterationUsage.usage.totalTokens
-				};
-				else if (iterationUsage.state === "invalid") output.usage.contextUsage = { state: "unavailable" };
-				else if (outputTokens !== void 0 && (messageStartPromptUsage !== void 0 || inputTokens !== void 0 && cacheReadTokens !== void 0 && cacheWriteTokens !== void 0)) {
-					const promptTokens = output.usage.input + output.usage.cacheRead + output.usage.cacheWrite;
-					output.usage.contextUsage = {
+				if (event.usage) {
+					const inputTokens = readAnthropicUsageTokenCount(event.usage.input_tokens);
+					if (inputTokens !== void 0) output.usage.input = inputTokens;
+					const outputTokens = readAnthropicUsageTokenCount(event.usage.output_tokens);
+					if (outputTokens !== void 0) output.usage.output = outputTokens;
+					const cacheReadTokens = readAnthropicUsageTokenCount(event.usage.cache_read_input_tokens);
+					if (cacheReadTokens !== void 0) output.usage.cacheRead = cacheReadTokens;
+					const cacheWriteTokens = readAnthropicUsageTokenCount(event.usage.cache_creation_input_tokens);
+					if (cacheWriteTokens !== void 0) output.usage.cacheWrite = cacheWriteTokens;
+					const cacheWriteUsage = readAnthropicCacheWriteUsage(event.usage);
+					if (cacheWriteUsage.cacheWrite1h !== void 0) output.usage.cacheWrite1h = cacheWriteUsage.cacheWrite1h;
+					output.usage.totalTokens = output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
+					const iterationUsage = readLastAnthropicIterationUsage(event.usage);
+					if (iterationUsage.state === "valid") output.usage.contextUsage = {
 						state: "available",
-						promptTokens,
-						totalTokens: promptTokens + output.usage.output
+						promptTokens: iterationUsage.usage.contextPromptTokens,
+						totalTokens: iterationUsage.usage.totalTokens
 					};
-				} else output.usage.contextUsage = { state: "unavailable" };
+					else if (iterationUsage.state === "invalid") output.usage.contextUsage = { state: "unavailable" };
+					else if (outputTokens !== void 0 && (messageStartPromptUsage !== void 0 || inputTokens !== void 0 && cacheReadTokens !== void 0 && cacheWriteTokens !== void 0)) {
+						const promptTokens = output.usage.input + output.usage.cacheRead + output.usage.cacheWrite;
+						output.usage.contextUsage = {
+							state: "available",
+							promptTokens,
+							totalTokens: promptTokens + output.usage.output
+						};
+					} else output.usage.contextUsage = { state: "unavailable" };
+				}
 				calculateCost(costModel, output.usage);
 			}
 			if (requestOptions?.signal?.aborted) throw new Error("Request was aborted");
@@ -842,6 +859,7 @@ const streamSimpleAnthropic = (model, context, options) => {
 	if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
 	const base = {
 		...buildBaseOptions(model, options, apiKey),
+		maxTokens: clampMaxTokensToModel(model, options?.maxTokens ?? model.maxTokens),
 		toolChoice: options?.toolChoice
 	};
 	const mandatoryAdaptiveThinking = requiresClaudeAdaptiveThinking(model);
@@ -869,10 +887,11 @@ const streamSimpleAnthropic = (model, context, options) => {
 		});
 	}
 	const adjusted = adjustMaxTokensForThinking(base.maxTokens, model.maxTokens, reasoning, options?.thinkingBudgets);
-	const thinkingEnabled = adjusted.thinkingBudget >= 1024;
+	const thinkingEnabled = adjusted.thinkingBudget >= ANTHROPIC_MIN_THINKING_BUDGET_TOKENS;
+	const maxTokens = thinkingEnabled ? adjusted.maxTokens : clampMaxTokensToModel(model, options?.maxTokens ?? model.maxTokens);
 	return streamAnthropic(model, context, {
 		...base,
-		maxTokens: adjusted.maxTokens,
+		maxTokens,
 		thinkingEnabled,
 		thinkingBudgetTokens: thinkingEnabled ? adjusted.thinkingBudget : void 0
 	});
@@ -1003,8 +1022,8 @@ function buildParams(model, context, isOAuthTokenResult, options, serverSideFall
 	const replayThinkingEnabled = mandatoryAdaptiveThinking || options?.thinkingEnabled === true;
 	const { cacheControl } = getCacheControl(model, options?.cacheRetention);
 	const system = buildAnthropicSystemBlocks(context.systemPrompt, isOAuthTokenResult, cacheControl);
-	const compat = context.tools ? getAnthropicCompat(model) : void 0;
-	const convertedTools = context.tools && compat ? convertTools(context.tools, isOAuthTokenResult, compat.supportsEagerToolInputStreaming, compat.supportsCacheControlOnTools ? cacheControl : void 0) : void 0;
+	const compat = getAnthropicCompat(model);
+	const convertedTools = context.tools ? convertTools(context.tools, isOAuthTokenResult, compat.supportsEagerToolInputStreaming, compat.supportsCacheControlOnTools ? cacheControl : void 0) : void 0;
 	const tools = convertedTools?.tools;
 	const toolProjection = convertedTools?.projection;
 	const systemCacheControlCount = countNativeCacheControlMarkers(system);
@@ -1012,7 +1031,7 @@ function buildParams(model, context, isOAuthTokenResult, options, serverSideFall
 	const messageCacheControlLimit = Math.max(0, ANTHROPIC_CACHE_CONTROL_LIMIT - systemCacheControlCount - toolCacheControlCount);
 	const params = {
 		model: model.id,
-		messages: convertMessages(context.messages, model, isOAuthTokenResult, cacheControl, messageCacheControlLimit, replayThinkingEnabled),
+		messages: convertMessages(context.messages, model, isOAuthTokenResult, cacheControl, messageCacheControlLimit, replayThinkingEnabled, compat.allowEmptySignature),
 		max_tokens: options?.maxTokens ?? model.maxTokens,
 		stream: true
 	};
@@ -1055,7 +1074,7 @@ function buildParams(model, context, isOAuthTokenResult, options, serverSideFall
 function normalizeToolCallId(id) {
 	return id.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
 }
-function convertMessages(messages, model, isOAuthTokenValue, cacheControl, messageCacheControlLimit = 4, replayThinkingEnabled = true) {
+function convertMessages(messages, model, isOAuthTokenValue, cacheControl, messageCacheControlLimit = 4, replayThinkingEnabled = true, allowEmptySignature = false) {
 	const params = [];
 	const cacheBreakpointOptOutParamIndexes = /* @__PURE__ */ new Set();
 	const transformedMessages = transformMessages(messages, model, normalizeToolCallId);
@@ -1123,7 +1142,7 @@ function convertMessages(messages, model, isOAuthTokenValue, cacheControl, messa
 				const thinkingSignature = block.thinkingSignature?.trim();
 				const hasNativeThinkingSignature = Boolean(thinkingSignature) && thinkingSignature !== "reasoning_content";
 				if (block.thinking.trim().length === 0 && !hasNativeThinkingSignature) continue;
-				if (!thinkingSignature) blocks.push({
+				if (!thinkingSignature && !allowEmptySignature) blocks.push({
 					type: "text",
 					text: sanitizeSurrogates(block.thinking)
 				});
@@ -1132,7 +1151,7 @@ function convertMessages(messages, model, isOAuthTokenValue, cacheControl, messa
 					blocks.push({
 						type: "thinking",
 						thinking: block.thinking,
-						signature: thinkingSignature
+						signature: thinkingSignature ?? ""
 					});
 				}
 			} else if (block.type === "toolCall") blocks.push({
@@ -1302,4 +1321,4 @@ function mapStopReason(reason) {
 	}
 }
 //#endregion
-export { readAnthropicFallbackBoundary as _, readAnthropicUsageTokenCount as a, usesFoundryBearerAuth as b, reconcileAnthropicToolChoice as c, findActiveAnthropicToolTurnAssistantIndex as d, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA as f, buildAnthropicServerSideFallbacks as g, applyAnthropicFallbackBoundary as h, readAnthropicPromptUsageSnapshot as i, resolveOriginalAnthropicToolName as l, CLAUDE_FABLE_5_FALLBACK_MODEL_COST as m, streamAnthropic as n, readLastAnthropicIterationUsage as o, CLAUDE_FABLE_5_FALLBACK_MODEL as p, streamSimpleAnthropic as r, projectAnthropicTools as s, anthropic_exports as t, ANTHROPIC_OMITTED_REASONING_TEXT as u, applyAnthropicRefusal as v, omitFoundryBearerCredentialHeaders as y };
+export { buildAnthropicServerSideFallbacks as _, readAnthropicPromptUsageSnapshot as a, omitFoundryBearerCredentialHeaders as b, projectAnthropicTools as c, ANTHROPIC_OMITTED_REASONING_TEXT as d, findActiveAnthropicToolTurnAssistantIndex as f, applyAnthropicFallbackBoundary as g, CLAUDE_FABLE_5_FALLBACK_MODEL_COST as h, readAnthropicCacheWriteUsage as i, reconcileAnthropicToolChoice as l, CLAUDE_FABLE_5_FALLBACK_MODEL as m, streamAnthropic as n, readAnthropicUsageTokenCount as o, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA as p, streamSimpleAnthropic as r, readLastAnthropicIterationUsage as s, anthropic_exports as t, resolveOriginalAnthropicToolName as u, readAnthropicFallbackBoundary as v, usesFoundryBearerAuth as x, applyAnthropicRefusal as y };

package/dist/{api-registry-Byvoz8Ha.d.mts → api-registry-DXBIXISf.d.mts}

@@ -1,4 +1,4 @@
-import { E as Model, F as SimpleStreamOptions, R as StreamFunction, n as Api, o as AssistantMessageEventStreamContract, u as Context, z as StreamOptions } from "./types-Cx2zJtyz.mjs";
+import { E as Model, F as SimpleStreamOptions, R as StreamFunction, n as Api, o as AssistantMessageEventStreamContract, u as Context, z as StreamOptions } from "./types-CYORveZ7.mjs";
 
 //#region packages/ai/src/api-registry.d.ts
 /** Runtime stream adapter signature stored in the API provider registry. */

package/dist/{azure-openai-responses-CHi7Wo4A.mjs → azure-openai-responses-BkZgyZ3I.mjs}

@@ -1,10 +1,10 @@
 import { n as getEnvApiKey } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
-import { l as buildBaseOptions } from "./transform-messages-BmS70CP5.mjs";
-import { a as resolveResponsesReasoningEffort, n as convertResponsesMessages, o as runResponsesStreamLifecycle, r as createResponsesAssistantOutput, t as applyCommonResponsesParams } from "./openai-responses-shared-Befb4D6R.mjs";
+import { l as buildBaseOptions } from "./transform-messages-eb6lxDLB.mjs";
+import { a as resolveResponsesReasoningEffort, n as convertResponsesMessages, o as runResponsesStreamLifecycle, r as createResponsesAssistantOutput, t as applyCommonResponsesParams } from "./openai-responses-shared-CwnGU0KW.mjs";
 import { i as resolveAzureDeploymentNameFromMap, t as isOpenAICompatibleAzureResponsesBaseUrl } from "./azure-openai-responses-client-compat-C7K7QfUE.mjs";
-import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-CFqm42J2.mjs";
+import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-WHf2nuOI.mjs";
 import OpenAI, { AzureOpenAI } from "openai";
 //#region packages/ai/src/providers/azure-openai-responses.ts
 const DEFAULT_AZURE_API_VERSION = "v1";
@@ -127,7 +127,8 @@ function buildParams(model, context, options, deploymentName) {
 		model: deploymentName,
 		input: convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS),
 		stream: true,
-		prompt_cache_key: options?.cacheRetention === "none" ? void 0 : clampOpenAIPromptCacheKey(options?.promptCacheKey ?? options?.sessionId)
+		prompt_cache_key: options?.cacheRetention === "none" ? void 0 : clampOpenAIPromptCacheKey(options?.promptCacheKey ?? options?.sessionId),
+		store: false
 	};
 	applyCommonResponsesParams(params, model, context, options);
 	return params;

package/dist/{event-stream-uQspnL8S.d.mts → event-stream-BPemFrFW.d.mts}

@@ -1,4 +1,4 @@
-import { a as AssistantMessageEvent, i as AssistantMessage, o as AssistantMessageEventStreamContract } from "./types-Cx2zJtyz.mjs";
+import { a as AssistantMessageEvent, i as AssistantMessage, o as AssistantMessageEventStreamContract } from "./types-CYORveZ7.mjs";
 
 //#region packages/llm-core/src/utils/event-stream.d.ts
 /** Generic async-iterable event stream with a separately awaited final result. */

package/dist/event-stream.d.mts

@@ -1,2 +1,2 @@
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-uQspnL8S.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-BPemFrFW.mjs";
 export { AssistantMessageEventStream, EventStream, createAssistantMessageEventStream };

package/dist/{google-BmsqsNwR.mjs → google-C_biGbV7.mjs}

@@ -1,8 +1,8 @@
 import { n as getEnvApiKey } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-4t713IeR.mjs";
-import { l as buildBaseOptions } from "./transform-messages-BmS70CP5.mjs";
-import { a as runGoogleGenerateContentLifecycle, i as getDisabledGoogleThinkingConfig, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-Dx2aba47.mjs";
+import { l as buildBaseOptions } from "./transform-messages-eb6lxDLB.mjs";
+import { a as runGoogleGenerateContentLifecycle, i as getDisabledGoogleThinkingConfig, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-BK-NoHl8.mjs";
 import { GoogleGenAI } from "@google/genai";
 //#region packages/ai/src/providers/google.ts
 let toolCallCounter = 0;

package/dist/{google-shared-Dx2aba47.mjs → google-shared-BK-NoHl8.mjs}

@@ -1,6 +1,7 @@
-import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Q1LSRIdo.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-1GiZ2_rr.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { C as stripSystemPromptCacheBoundary, i as extractToolResultText, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, t as transformMessages } from "./transform-messages-BmS70CP5.mjs";
+import { i as extractToolResultText, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, t as transformMessages, w as stripSystemPromptCacheBoundary } from "./transform-messages-eb6lxDLB.mjs";
+import { t as formatProviderError } from "./provider-error-LMfTEkfO.mjs";
 import { FinishReason, FunctionCallingConfigMode, ThinkingLevel } from "@google/genai";
 //#region packages/ai/src/providers/google-shared.ts
 /**
@@ -273,7 +274,7 @@ async function runGoogleGenerateContentLifecycle(params) {
 	} catch (error) {
 		for (const block of output.content) if ("index" in block) delete block.index;
 		output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-		output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+		output.errorMessage = formatProviderError(error);
 		stream.push({
 			type: "error",
 			reason: output.stopReason,

package/dist/{google-vertex-HZ_0rTwS.mjs → google-vertex-CvdzF1U7.mjs}

@@ -1,7 +1,7 @@
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-4t713IeR.mjs";
-import { l as buildBaseOptions } from "./transform-messages-BmS70CP5.mjs";
-import { a as runGoogleGenerateContentLifecycle, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-Dx2aba47.mjs";
+import { l as buildBaseOptions } from "./transform-messages-eb6lxDLB.mjs";
+import { a as runGoogleGenerateContentLifecycle, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-BK-NoHl8.mjs";
 import { GoogleGenAI, ResourceScope } from "@google/genai";
 //#region packages/ai/src/providers/google-vertex.ts
 const API_VERSION = "v1";

package/dist/index.d.mts

@@ -1,9 +1,9 @@
 import { a as resolveClaudeFable5ModelIdentity, c as resolveClaudeNativeThinkingLevelMap, d as supportsClaudeNativeMaxEffort, f as supportsClaudeNativeXhighEffort, i as requiresClaudeMandatoryAdaptiveThinking, l as resolveClaudeSonnet5ModelIdentity, n as CLAUDE_SONNET_5_THINKING_PROFILE, o as resolveClaudeModelIdentity, r as requiresClaudeDefaultSampling, s as resolveClaudeMythos5ModelIdentity, t as CLAUDE_FABLE_5_THINKING_PROFILE, u as supportsClaudeAdaptiveThinking } from "./index-BoTnz8cv.mjs";
 import { a as extractDiagnosticError, i as createAssistantMessageDiagnostic, n as DiagnosticErrorInfo, o as formatThrownValue, r as appendAssistantMessageDiagnostic, t as AssistantMessageDiagnostic } from "./diagnostics-BaTA9eVl.mjs";
-import { $ as VercelGatewayRouting, A as OpenRouterRouting, B as TextContent, C as KnownImagesProvider, D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, G as ThinkingLevelMap, H as ThinkingBudgets, I as StopReason, J as ToolResultMessage, K as Tool, L as StreamFn, M as ProviderImagesOptions, N as ProviderResponse, O as OpenAICompletionsCompat, P as ProviderStreamOptions, Q as ValidateToolArgumentsFn, R as StreamFunction, S as KnownImagesApi, T as Message, U as ThinkingContent, V as TextSignatureV1, W as ThinkingLevel, X as Usage, Y as Transport, Z as UserMessage, _ as ImagesOptions, a as AssistantMessageEvent, b as ImagesStopReason, c as CacheRetention, d as ImageContent, f as ImagesApi, g as ImagesModel, h as ImagesInputContent, i as AssistantMessage, j as Provider, k as OpenAIResponsesCompat, l as CompleteSimpleFn, m as ImagesFunction, n as Api, o as AssistantMessageEventStreamContract, p as ImagesContext, q as ToolCall, r as AssistantImages, s as AssistantMessageEventStreamLike, t as AnthropicMessagesCompat, u as Context, v as ImagesOutputContent, w as MaybePromise, x as KnownApi, y as ImagesProvider, z as StreamOptions } from "./types-Cx2zJtyz.mjs";
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-uQspnL8S.mjs";
-import { n as validateToolCall, t as validateToolArguments } from "./validation-Cej7htKc.mjs";
-import { a as RegisteredApiProvider, i as ApiStreamSimpleFunction, n as ApiRegistry, o as createApiRegistry, r as ApiStreamFunction, t as ApiProvider } from "./api-registry-Byvoz8Ha.mjs";
+import { $ as VercelGatewayRouting, A as OpenRouterRouting, B as TextContent, C as KnownImagesProvider, D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, G as ThinkingLevelMap, H as ThinkingBudgets, I as StopReason, J as ToolResultMessage, K as Tool, L as StreamFn, M as ProviderImagesOptions, N as ProviderResponse, O as OpenAICompletionsCompat, P as ProviderStreamOptions, Q as ValidateToolArgumentsFn, R as StreamFunction, S as KnownImagesApi, T as Message, U as ThinkingContent, V as TextSignatureV1, W as ThinkingLevel, X as Usage, Y as Transport, Z as UserMessage, _ as ImagesOptions, a as AssistantMessageEvent, b as ImagesStopReason, c as CacheRetention, d as ImageContent, f as ImagesApi, g as ImagesModel, h as ImagesInputContent, i as AssistantMessage, j as Provider, k as OpenAIResponsesCompat, l as CompleteSimpleFn, m as ImagesFunction, n as Api, o as AssistantMessageEventStreamContract, p as ImagesContext, q as ToolCall, r as AssistantImages, s as AssistantMessageEventStreamLike, t as AnthropicMessagesCompat, u as Context, v as ImagesOutputContent, w as MaybePromise, x as KnownApi, y as ImagesProvider, z as StreamOptions } from "./types-CYORveZ7.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-BPemFrFW.mjs";
+import { n as validateToolCall, t as validateToolArguments } from "./validation-BX8euiDv.mjs";
+import { a as RegisteredApiProvider, i as ApiStreamSimpleFunction, n as ApiRegistry, o as createApiRegistry, r as ApiStreamFunction, t as ApiProvider } from "./api-registry-DXBIXISf.mjs";
 
 //#region packages/ai/src/host.d.ts
 /** Strict-tool policy inputs for OpenAI-compatible routes. */

package/dist/internal/anthropic.d.mts

@@ -1,6 +1,6 @@
 import { a as resolveClaudeFable5ModelIdentity, c as resolveClaudeNativeThinkingLevelMap, d as supportsClaudeNativeMaxEffort, f as supportsClaudeNativeXhighEffort, i as requiresClaudeMandatoryAdaptiveThinking, l as resolveClaudeSonnet5ModelIdentity, o as resolveClaudeModelIdentity, r as requiresClaudeDefaultSampling, s as resolveClaudeMythos5ModelIdentity, u as supportsClaudeAdaptiveThinking } from "../index-BoTnz8cv.mjs";
 import { t as AssistantMessageDiagnostic } from "../diagnostics-BaTA9eVl.mjs";
-import { E as Model, F as SimpleStreamOptions, R as StreamFunction, u as Context, z as StreamOptions } from "../types-Cx2zJtyz.mjs";
+import { E as Model, F as SimpleStreamOptions, R as StreamFunction, u as Context, z as StreamOptions } from "../types-CYORveZ7.mjs";
 import Anthropic from "@anthropic-ai/sdk";
 
 //#region packages/ai/src/providers/anthropic.d.ts
@@ -208,8 +208,13 @@ type AnthropicUsagePayload = {
   output_tokens?: unknown;
   cache_read_input_tokens?: unknown;
   cache_creation_input_tokens?: unknown;
+  cache_creation?: unknown;
   iterations?: unknown;
 };
+type AnthropicCacheWriteUsage = {
+  cacheWrite5m?: number;
+  cacheWrite1h?: number;
+};
 type AnthropicPromptUsageSnapshot = {
   input: number;
   cacheRead: number;
@@ -228,7 +233,8 @@ type AnthropicIterationUsageResult = {
   usage: AnthropicIterationUsageSnapshot;
 };
 declare function readAnthropicUsageTokenCount(value: unknown): number | undefined;
+declare function readAnthropicCacheWriteUsage(usage: AnthropicUsagePayload): AnthropicCacheWriteUsage;
 declare function readAnthropicPromptUsageSnapshot(usage: AnthropicUsagePayload): AnthropicPromptUsageSnapshot | undefined;
 declare function readLastAnthropicIterationUsage(usage: AnthropicUsagePayload): AnthropicIterationUsageResult;
 //#endregion
-export { ANTHROPIC_OMITTED_REASONING_TEXT, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, AnthropicEffort, AnthropicFallbackBoundary, AnthropicIterationUsageResult, AnthropicIterationUsageSnapshot, AnthropicOptions, AnthropicProjectedToolChoice, AnthropicPromptUsageSnapshot, AnthropicThinkingDisplay, AnthropicToolProjection, CLAUDE_FABLE_5_FALLBACK_MODEL, CLAUDE_FABLE_5_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, buildAnthropicServerSideFallbacks, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeSonnet5RequestContext, projectAnthropicTools, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };
+export { ANTHROPIC_OMITTED_REASONING_TEXT, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, AnthropicCacheWriteUsage, AnthropicEffort, AnthropicFallbackBoundary, AnthropicIterationUsageResult, AnthropicIterationUsageSnapshot, AnthropicOptions, AnthropicProjectedToolChoice, AnthropicPromptUsageSnapshot, AnthropicThinkingDisplay, AnthropicToolProjection, CLAUDE_FABLE_5_FALLBACK_MODEL, CLAUDE_FABLE_5_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, buildAnthropicServerSideFallbacks, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeSonnet5RequestContext, projectAnthropicTools, readAnthropicCacheWriteUsage, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };

package/dist/internal/anthropic.mjs

@@ -1,4 +1,4 @@
 import { a as resolveClaudeFable5ModelIdentity, c as resolveClaudeNativeThinkingLevelMap, d as supportsClaudeNativeMaxEffort, f as supportsClaudeNativeXhighEffort, i as requiresClaudeMandatoryAdaptiveThinking, l as resolveClaudeSonnet5ModelIdentity, o as resolveClaudeModelIdentity, r as requiresClaudeDefaultSampling, s as resolveClaudeMythos5ModelIdentity, u as supportsClaudeAdaptiveThinking } from "../src-CXno1H5g.mjs";
-import { _ as usesClaudeFable5MessagesContract, f as applyClaudeRequestContract, g as resolveModelBoundThinkingReplayMode, h as requiresClaudeAdaptiveThinking, m as prepareClaudeSonnet5RequestContext, p as defaultsClaudeAdaptiveThinking, v as usesClaudeStreamingRefusalContract } from "../transform-messages-BmS70CP5.mjs";
-import { _ as readAnthropicFallbackBoundary, a as readAnthropicUsageTokenCount, b as usesFoundryBearerAuth, c as reconcileAnthropicToolChoice, d as findActiveAnthropicToolTurnAssistantIndex, f as ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, g as buildAnthropicServerSideFallbacks, h as applyAnthropicFallbackBoundary, i as readAnthropicPromptUsageSnapshot, l as resolveOriginalAnthropicToolName, m as CLAUDE_FABLE_5_FALLBACK_MODEL_COST, n as streamAnthropic, o as readLastAnthropicIterationUsage, p as CLAUDE_FABLE_5_FALLBACK_MODEL, r as streamSimpleAnthropic, s as projectAnthropicTools, u as ANTHROPIC_OMITTED_REASONING_TEXT, v as applyAnthropicRefusal, y as omitFoundryBearerCredentialHeaders } from "../anthropic-BIRSg5x9.mjs";
-export { ANTHROPIC_OMITTED_REASONING_TEXT, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, CLAUDE_FABLE_5_FALLBACK_MODEL, CLAUDE_FABLE_5_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, buildAnthropicServerSideFallbacks, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeSonnet5RequestContext, projectAnthropicTools, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };
+import { _ as resolveModelBoundThinkingReplayMode, g as requiresClaudeAdaptiveThinking, h as prepareClaudeSonnet5RequestContext, m as defaultsClaudeAdaptiveThinking, p as applyClaudeRequestContract, v as usesClaudeFable5MessagesContract, y as usesClaudeStreamingRefusalContract } from "../transform-messages-eb6lxDLB.mjs";
+import { _ as buildAnthropicServerSideFallbacks, a as readAnthropicPromptUsageSnapshot, b as omitFoundryBearerCredentialHeaders, c as projectAnthropicTools, d as ANTHROPIC_OMITTED_REASONING_TEXT, f as findActiveAnthropicToolTurnAssistantIndex, g as applyAnthropicFallbackBoundary, h as CLAUDE_FABLE_5_FALLBACK_MODEL_COST, i as readAnthropicCacheWriteUsage, l as reconcileAnthropicToolChoice, m as CLAUDE_FABLE_5_FALLBACK_MODEL, n as streamAnthropic, o as readAnthropicUsageTokenCount, p as ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, r as streamSimpleAnthropic, s as readLastAnthropicIterationUsage, u as resolveOriginalAnthropicToolName, v as readAnthropicFallbackBoundary, x as usesFoundryBearerAuth, y as applyAnthropicRefusal } from "../anthropic-D7SyZQ5v.mjs";
+export { ANTHROPIC_OMITTED_REASONING_TEXT, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, CLAUDE_FABLE_5_FALLBACK_MODEL, CLAUDE_FABLE_5_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, buildAnthropicServerSideFallbacks, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeSonnet5RequestContext, projectAnthropicTools, readAnthropicCacheWriteUsage, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };

package/dist/internal/openai.d.mts

@@ -1,4 +1,4 @@
-import { E as Model, F as SimpleStreamOptions, I as StopReason, O as OpenAICompletionsCompat, R as StreamFunction, u as Context, z as StreamOptions } from "../types-Cx2zJtyz.mjs";
+import { E as Model, F as SimpleStreamOptions, I as StopReason, O as OpenAICompletionsCompat, R as StreamFunction, u as Context, z as StreamOptions } from "../types-CYORveZ7.mjs";
 import OpenAI from "openai";
 import { TSchema } from "typebox";
 import { ChatCompletionMessageParam } from "openai/resources/chat/completions.js";
@@ -99,6 +99,7 @@ interface OpenAICompletionsOptions extends StreamOptions {
 }
 type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "cacheControlFormat"> & {
   cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];
+  sessionAffinityFormat: "openai" | "openrouter";
 };
 declare const streamOpenAICompletions: StreamFunction<"openai-completions", OpenAICompletionsOptions>;
 declare const streamSimpleOpenAICompletions: StreamFunction<"openai-completions", SimpleStreamOptions>;

package/dist/internal/openai.mjs

@@ -1,7 +1,7 @@
 import { t as projectRuntimeToolInputSchema } from "../tool-schema-json-projection-BwNu3nDi.mjs";
-import { A as resolveOpenAISupportedReasoningEfforts, C as isResponsesTextDeltaEventType, D as isOpenAIGpt56Model, E as isOpenAIGpt55Model, F as resolveUnsupportedToolSchemaKeywords, I as shouldOmitEmptyArrayItems, L as stripUnsupportedSchemaKeywords, M as supportsOpenAITemperature, N as extractToolSchemaModelCompat, O as normalizeOpenAIReasoningEffort, P as normalizeToolParameterSchema, R as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, S as isResponsesTextContentPartType, T as isOpenAIGpt54MiniModel, _ as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, b as isAzureResponsesTextDeltaEvent, c as clearOpenAIToolSchemaCacheForTest, d as normalizeOpenAIStrictToolParameters, f as normalizeStrictOpenAIJsonSchema, g as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, h as readResponsesToolCallItemIdentity, j as supportsOpenAIReasoningEffort, k as resolveOpenAIReasoningEffortForModel, l as findOpenAIStrictToolProjectionDiagnostics, m as createResponsesToolCallTracker, p as resolveOpenAIProjectedToolsStrictToolFlag, u as isStrictOpenAIJsonSchemaCompatible, v as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, w as resolveResponsesMessageSnapshotCollapse, x as isAzureResponsesTextDeltaEventType, y as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, z as cleanSchemaForGemini } from "../openai-responses-shared-Befb4D6R.mjs";
+import { A as resolveOpenAISupportedReasoningEfforts, C as isResponsesTextDeltaEventType, D as isOpenAIGpt56Model, E as isOpenAIGpt55Model, F as resolveUnsupportedToolSchemaKeywords, I as shouldOmitEmptyArrayItems, L as stripUnsupportedSchemaKeywords, M as supportsOpenAITemperature, N as extractToolSchemaModelCompat, O as normalizeOpenAIReasoningEffort, P as normalizeToolParameterSchema, R as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, S as isResponsesTextContentPartType, T as isOpenAIGpt54MiniModel, _ as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, b as isAzureResponsesTextDeltaEvent, c as clearOpenAIToolSchemaCacheForTest, d as normalizeOpenAIStrictToolParameters, f as normalizeStrictOpenAIJsonSchema, g as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, h as readResponsesToolCallItemIdentity, j as supportsOpenAIReasoningEffort, k as resolveOpenAIReasoningEffortForModel, l as findOpenAIStrictToolProjectionDiagnostics, m as createResponsesToolCallTracker, p as resolveOpenAIProjectedToolsStrictToolFlag, u as isStrictOpenAIJsonSchemaCompatible, v as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, w as resolveResponsesMessageSnapshotCollapse, x as isAzureResponsesTextDeltaEventType, y as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, z as cleanSchemaForGemini } from "../openai-responses-shared-CwnGU0KW.mjs";
 import { i as resolveAzureDeploymentNameFromMap, n as isTraditionalAzureOpenAIHost, r as parseAzureDeploymentNameMap, t as isOpenAICompatibleAzureResponsesBaseUrl } from "../azure-openai-responses-client-compat-C7K7QfUE.mjs";
-import { a as clampOpenAIPromptCacheKey, i as OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, n as reconcileOpenAICompletionsToolChoice, r as reconcileOpenAIResponsesToolChoice, t as projectOpenAITools } from "../openai-tool-projection-CFqm42J2.mjs";
-import { a as mapOpenAIStopReason, i as streamSimpleOpenAICompletions, r as streamOpenAICompletions, t as convertMessages } from "../openai-completions-DPR_O2RW.mjs";
-import { n as streamOpenAIResponses, r as streamSimpleOpenAIResponses } from "../openai-responses-Cov-Vdc2.mjs";
+import { a as clampOpenAIPromptCacheKey, i as OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, n as reconcileOpenAICompletionsToolChoice, r as reconcileOpenAIResponsesToolChoice, t as projectOpenAITools } from "../openai-tool-projection-WHf2nuOI.mjs";
+import { a as mapOpenAIStopReason, i as streamSimpleOpenAICompletions, r as streamOpenAICompletions, t as convertMessages } from "../openai-completions-zLDfebnN.mjs";
+import { n as streamOpenAIResponses, r as streamSimpleOpenAIResponses } from "../openai-responses-ClspWt4I.mjs";
 export { AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, clampOpenAIPromptCacheKey, cleanSchemaForGemini, clearOpenAIToolSchemaCacheForTest, convertMessages, createResponsesToolCallTracker, extractToolSchemaModelCompat, findOpenAIStrictToolProjectionDiagnostics, isAzureResponsesTextDeltaEvent, isAzureResponsesTextDeltaEventType, isOpenAICompatibleAzureResponsesBaseUrl, isOpenAIGpt54MiniModel, isOpenAIGpt55Model, isOpenAIGpt56Model, isResponsesTextContentPartType, isResponsesTextDeltaEventType, isStrictOpenAIJsonSchemaCompatible, isTraditionalAzureOpenAIHost, mapOpenAIStopReason, normalizeOpenAIReasoningEffort, normalizeOpenAIStrictToolParameters, normalizeStrictOpenAIJsonSchema, normalizeToolParameterSchema, parseAzureDeploymentNameMap, projectOpenAITools, projectRuntimeToolInputSchema, readResponsesToolCallItemIdentity, reconcileOpenAICompletionsToolChoice, reconcileOpenAIResponsesToolChoice, resolveAzureDeploymentNameFromMap, resolveOpenAIProjectedToolsStrictToolFlag, resolveOpenAIReasoningEffortForModel, resolveOpenAISupportedReasoningEfforts, resolveResponsesMessageSnapshotCollapse, resolveUnsupportedToolSchemaKeywords, shouldOmitEmptyArrayItems, streamOpenAICompletions, streamOpenAIResponses, streamSimpleOpenAICompletions, streamSimpleOpenAIResponses, stripUnsupportedSchemaKeywords, supportsOpenAIReasoningEffort, supportsOpenAITemperature };

package/dist/internal/runtime.d.mts

@@ -1,5 +1,5 @@
-import { D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, P as ProviderStreamOptions, X as Usage, i as AssistantMessage, n as Api, o as AssistantMessageEventStreamContract, u as Context, z as StreamOptions } from "../types-Cx2zJtyz.mjs";
-import { a as RegisteredApiProvider, t as ApiProvider } from "../api-registry-Byvoz8Ha.mjs";
+import { D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, P as ProviderStreamOptions, X as Usage, i as AssistantMessage, n as Api, o as AssistantMessageEventStreamContract, u as Context, z as StreamOptions } from "../types-CYORveZ7.mjs";
+import { a as RegisteredApiProvider, t as ApiProvider } from "../api-registry-DXBIXISf.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-BZiVbGwK.mjs";
 
 //#region packages/ai/src/internal/default-runtime.d.ts

package/dist/internal/runtime.mjs

@@ -1,6 +1,6 @@
 import { n as getEnvApiKey, t as findEnvKeys } from "../env-api-keys-DnhYpc27.mjs";
 import { n as createApiRegistry, t as createLlmRuntime } from "../stream-CREqxHgU.mjs";
-import { a as modelsAreEqual, i as getSupportedThinkingLevels, n as calculateCost, r as clampThinkingLevel, t as applyProviderReportedUsageCost } from "../model-utils-Q1LSRIdo.mjs";
+import { a as modelsAreEqual, i as getSupportedThinkingLevels, n as calculateCost, r as clampThinkingLevel, t as applyProviderReportedUsageCost } from "../model-utils-1GiZ2_rr.mjs";
 import { n as onLlmRequestActivity, r as createDeferredEventBuffer, t as notifyLlmRequestActivity } from "../llm-request-activity-CehVkZP-.mjs";
 import { t as headersToRecord } from "../headers-B_e4-1J0.mjs";
 import { n as parseStreamingJson, r as repairJson, t as parseJsonWithRepair } from "../json-parse-BvXNt1-7.mjs";
@@ -70,11 +70,12 @@ const OVERFLOW_PATTERNS = [
 	/request_too_large/i,
 	/input is too long for requested model/i,
 	/exceeds the context window/i,
-	/exceeds (?:the )?(?:model'?s )?maximum context length of [\d,]+ tokens?/i,
+	/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i,
 	/input token count.*exceeds the maximum/i,
 	/maximum prompt length is \d+/i,
 	/reduce the length of the messages/i,
 	/maximum context length is \d+ tokens/i,
+	/exceeds (?:the )?maximum allowed input length of [\d,]+ tokens?/i,
 	/input \(\d+ tokens\) is longer than the model'?s context length \(\d+ tokens\)/i,
 	/exceeds the limit of \d+/i,
 	/exceeds the available context size/i,

package/dist/internal/shared.d.mts

@@ -1,4 +1,4 @@
-import { E as Model, F as SimpleStreamOptions, H as ThinkingBudgets, T as Message, W as ThinkingLevel, i as AssistantMessage, n as Api, z as StreamOptions } from "../types-Cx2zJtyz.mjs";
+import { E as Model, F as SimpleStreamOptions, H as ThinkingBudgets, T as Message, W as ThinkingLevel, i as AssistantMessage, n as Api, z as StreamOptions } from "../types-CYORveZ7.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-BZiVbGwK.mjs";
 
 //#region packages/ai/src/providers/simple-options.d.ts
@@ -7,6 +7,8 @@ type FirstEventStreamOptions = {
   onFirstEventTimeout?: (reason: Error) => void;
 };
 declare function buildBaseOptions(model: Model, options?: SimpleStreamOptions, apiKey?: string): StreamOptions & FirstEventStreamOptions;
+declare function clampMaxTokensToModel(model: Model, requestedMaxTokens: number): number;
+declare function clampMaxTokensToModel(model: Model, requestedMaxTokens: number | undefined): number | undefined;
 declare function clampReasoning(effort: ThinkingLevel): Exclude<ThinkingLevel, "xhigh">;
 declare function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh"> | undefined;
 declare function adjustMaxTokensForThinking(baseMaxTokens: number | undefined, modelMaxTokens: number, reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): {
@@ -55,4 +57,4 @@ declare function prependSystemPromptAdditionAfterCacheBoundary(params: {
   systemPromptAddition?: string;
 }): string;
 //#endregion
-export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };
+export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToModel, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };

package/dist/internal/shared.mjs

@@ -1,3 +1,3 @@
 import { t as sanitizeSurrogates } from "../sanitize-unicode-DT5o51ur.mjs";
-import { C as stripSystemPromptCacheBoundary, S as splitSystemPromptCacheBoundary, T as normalizeStructuredPromptSection, a as hasMediaPayload, b as ensureSystemPromptCacheBoundary, c as adjustMaxTokensForThinking, i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, r as extractToolResultBlockText, t as transformMessages, u as clampReasoning, w as normalizePromptCapabilityIds, x as prependSystemPromptAdditionAfterCacheBoundary, y as SYSTEM_PROMPT_CACHE_BOUNDARY } from "../transform-messages-BmS70CP5.mjs";
-export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };
+import { C as splitSystemPromptCacheBoundary, E as normalizeStructuredPromptSection, S as prependSystemPromptAdditionAfterCacheBoundary, T as normalizePromptCapabilityIds, a as hasMediaPayload, b as SYSTEM_PROMPT_CACHE_BOUNDARY, c as adjustMaxTokensForThinking, d as clampReasoning, i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, r as extractToolResultBlockText, t as transformMessages, u as clampMaxTokensToModel, w as stripSystemPromptCacheBoundary, x as ensureSystemPromptCacheBoundary } from "../transform-messages-eb6lxDLB.mjs";
+export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToModel, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };

package/dist/{mistral-DxMPt9q9.mjs → mistral-BdrQmDWp.mjs}

@@ -1,10 +1,10 @@
 import { n as getEnvApiKey } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
-import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Q1LSRIdo.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-1GiZ2_rr.mjs";
 import { n as parseStreamingJson } from "./json-parse-BvXNt1-7.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { C as stripSystemPromptCacheBoundary, i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, s as truncateUtf16Safe, t as transformMessages } from "./transform-messages-BmS70CP5.mjs";
+import { i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, s as truncateUtf16Safe, t as transformMessages, u as clampMaxTokensToModel, w as stripSystemPromptCacheBoundary } from "./transform-messages-eb6lxDLB.mjs";
 import { t as shortHash } from "./hash-CHgqbJmD.mjs";
 import { t as createSseByteGuard } from "./streaming-byte-guard-BrbkbwUu.mjs";
 import { randomUUID } from "node:crypto";
@@ -75,7 +75,7 @@ const streamMistral = (model, context, options) => {
 				...model.headers,
 				...options?.headers
 			};
-			if (options?.sessionId) headers["x-affinity"] ||= options.sessionId;
+			if (resolveMistralPromptCacheKey(options) && options?.sessionId) headers["x-affinity"] ||= options.sessionId;
 			const mistralStream = await mistral.chat.stream(payload, {
 				headers,
 				signal: options?.signal
@@ -113,7 +113,10 @@ const streamMistral = (model, context, options) => {
 const streamSimpleMistral = (model, context, options) => {
 	const apiKey = options?.apiKey || getEnvApiKey(model.provider);
 	if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
-	const base = buildBaseOptions(model, options, apiKey);
+	const base = {
+		...buildBaseOptions(model, options, apiKey),
+		maxTokens: clampMaxTokensToModel(model, options?.maxTokens)
+	};
 	const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : void 0;
 	const reasoning = clampedReasoning === "off" ? void 0 : clampedReasoning;
 	const shouldUseReasoning = model.reasoning && reasoning !== void 0;
@@ -218,12 +221,23 @@ function buildChatPayload(model, context, messages, options) {
 	}
 	if (options?.promptMode) payload.promptMode = options.promptMode;
 	if (options?.reasoningEffort) payload.reasoningEffort = options.reasoningEffort;
+	const promptCacheKey = resolveMistralPromptCacheKey(options);
+	if (promptCacheKey) payload.promptCacheKey = promptCacheKey;
 	if (context.systemPrompt) payload.messages.unshift({
 		role: "system",
 		content: sanitizeSurrogates(stripSystemPromptCacheBoundary(context.systemPrompt))
 	});
 	return payload;
 }
+function resolveMistralPromptCacheKey(options) {
+	if (options?.cacheRetention === "none") return;
+	return options?.promptCacheKey?.trim() || options?.sessionId?.trim() || void 0;
+}
+function readMistralCachedPromptTokens(usage, promptTokens) {
+	const record = usage;
+	const rawCachedTokens = record.promptTokensDetails?.cachedTokens ?? record.prompt_tokens_details?.cached_tokens ?? record.cachedTokens ?? record.cached_tokens;
+	return Math.min(promptTokens, Math.max(0, typeof rawCachedTokens === "number" && Number.isFinite(rawCachedTokens) ? rawCachedTokens : 0));
+}
 async function consumeChatStream(model, output, stream, mistralStream) {
 	let currentBlock = null;
 	const blocks = output.content;
@@ -309,11 +323,13 @@ async function consumeChatStream(model, output, stream, mistralStream) {
 		const chunk = event.data;
 		output.responseId ||= chunk.id;
 		if (chunk.usage) {
-			output.usage.input = chunk.usage.promptTokens || 0;
+			const promptTokens = chunk.usage.promptTokens || 0;
+			const cachedPromptTokens = readMistralCachedPromptTokens(chunk.usage, promptTokens);
+			output.usage.input = Math.max(0, promptTokens - cachedPromptTokens);
 			output.usage.output = chunk.usage.completionTokens || 0;
-			output.usage.cacheRead = 0;
+			output.usage.cacheRead = cachedPromptTokens;
 			output.usage.cacheWrite = 0;
-			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;
+			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output + output.usage.cacheRead;
 			calculateCost(model, output.usage);
 		}
 		const choice = chunk.choices[0];

package/dist/{model-utils-Q1LSRIdo.mjs → model-utils-1GiZ2_rr.mjs}

@@ -2,10 +2,12 @@ import { c as resolveClaudeNativeThinkingLevelMap, i as requiresClaudeMandatoryA
 //#region packages/ai/src/model-utils.ts
 /** Calculates and stores model cost fields from token usage and per-million pricing. */
 function calculateCost(model, usage) {
+	const cacheWrite1h = Math.min(usage.cacheWrite, Math.max(0, usage.cacheWrite1h ?? 0));
+	const cacheWrite5m = usage.cacheWrite - cacheWrite1h;
 	usage.cost.input = model.cost.input / 1e6 * usage.input;
 	usage.cost.output = model.cost.output / 1e6 * usage.output;
 	usage.cost.cacheRead = model.cost.cacheRead / 1e6 * usage.cacheRead;
-	usage.cost.cacheWrite = model.cost.cacheWrite / 1e6 * usage.cacheWrite;
+	usage.cost.cacheWrite = (model.cost.cacheWrite * cacheWrite5m + model.cost.input * 2 * cacheWrite1h) / 1e6;
 	usage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;
 	return usage.cost;
 }