@openclaw/ai 2026.7.2-beta.7 → 2026.7.34

package/dist/internal/runtime.mjs

@@ -1,14 +1,13 @@
-import { n as getEnvApiKey, t as findEnvKeys } from "../env-api-keys-DrgeBuva.mjs";
+import { n as getEnvApiKey, t as findEnvKeys } from "../env-api-keys-CtMlqaQ4.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-Cn9KAz3z.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-DzNSIQBq.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-DT5o51ur.mjs";
-import { c as calculateCost, d as modelsAreEqual, l as clampThinkingLevel, s as applyProviderReportedUsageCost, u as getSupportedThinkingLevels } from "../number-coercion-DvG7SNMg.mjs";
-import { t as createDeferredEventBuffer } from "../deferred-event-buffer-DAvyP7qA.mjs";
-import { n as parseStreamingJson, r as repairJson, t as parseJsonWithRepair } from "../json-parse-BvXNt1-7.mjs";
-import { n as onLlmRequestActivity, t as notifyLlmRequestActivity } from "../llm-request-activity-BjtkplhG.mjs";
-import { t as createReasoningTagTextPartitioner } from "../reasoning-tag-text-partitioner-CGDyLWUR.mjs";
-import { a as withFirstStreamEventTimeout, i as getFirstStreamEventTimeoutMs, n as createFirstStreamEventTimeoutError, r as getFirstStreamEventTimeoutHandler, t as createFirstStreamEventAbortController } from "../stream-first-event-timeout-BBys9hSb.mjs";
+import { t as createReasoningTagTextPartitioner } from "../reasoning-tag-text-partitioner-axhAdUwg.mjs";
+import { a as withFirstStreamEventTimeout, i as getFirstStreamEventTimeoutMs, n as createFirstStreamEventTimeoutError, r as getFirstStreamEventTimeoutHandler, t as createFirstStreamEventAbortController } from "../stream-first-event-timeout-RjWszj8c.mjs";
 import { t as shortHash } from "../hash-CHgqbJmD.mjs";
-import { t as headersToRecord } from "../headers-B_e4-1J0.mjs";
 import { i as registerSessionResourceCleanup, n as resolveOpenAICodexAccountId, r as cleanupSessionResources, t as decodeOpenAICodexJwtPayload } from "../openai-chatgpt-jwt-DhAAzLkj.mjs";
 import { t as createSseByteGuard } from "../streaming-byte-guard-BrbkbwUu.mjs";
 //#region packages/ai/src/internal/default-runtime.ts
@@ -27,10 +26,7 @@ function resolveDefaultRuntime() {
 const defaultRuntime = resolveDefaultRuntime();
 const defaultApiRegistry = defaultRuntime.registry;
 const defaultLlmRuntime = defaultRuntime.runtime;
-const { getApiProvider, getApiProviders } = defaultApiRegistry;
-function clearApiProviders() {
-	defaultApiRegistry.clearApiProviders();
-}
+const { registerApiProvider, getApiProvider, getApiProviders, unregisterApiProviders, clearApiProviders } = defaultApiRegistry;
 const { stream, complete, streamSimple, completeSimple } = defaultLlmRuntime;
 //#endregion
 //#region packages/ai/src/utils/overflow.ts
@@ -63,9 +59,7 @@ function isConfiguredContextSizeOverflowError(errorMessage) {
 * - Kimi For Coding: "Your request exceeded model token limit: X (requested: Y)"
 * - Cerebras: "400/413 status code (no body)"
 * - Mistral: "Prompt contains X tokens ... too large for model with Y maximum context length"
-* - z.ai: May return "tokens in request more than max tokens allowed" (code 1210),
-*   "Prompt exceeds max length" (code 1261), or accept overflow silently; handled via the
-*   error patterns or usage.input > contextWindow
+* - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow
 * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason "length"
 *   with output=0 (no room left to generate). Detected via stopReason "length" + zero output +
 *   input filling the context window.
@@ -76,20 +70,17 @@ 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?|\s*\([\d,]+\))/i,
+	/exceeds (?:the )?(?:model'?s )?maximum context length of [\d,]+ tokens?/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,
 	/greater than the context length/i,
 	/context window exceeds limit/i,
 	/exceeded model token limit/i,
-	/tokens? in request more than max tokens? allowed/i,
-	/prompt exceeds max(?:imum)? length/i,
 	/too large for model with \d+ maximum context length/i,
 	CONFIGURED_CONTEXT_SIZE_OVERFLOW_RE,
 	/model_context_window_exceeded/i,
@@ -116,7 +107,7 @@ const NON_OVERFLOW_PATTERNS = [
 function resolveContextInputTokens(message) {
 	if (message.usage.contextUsage?.state === "available") return message.usage.contextUsage.promptTokens;
 	if (message.usage.contextUsage?.state === "unavailable") return;
-	return message.usage.input + message.usage.cacheRead + message.usage.cacheWrite;
+	return message.usage.input + message.usage.cacheRead;
 }
 /**
 * Check if an assistant message represents a context overflow error.
@@ -142,12 +133,10 @@ function resolveContextInputTokens(message) {
 * - llama.cpp: "exceeds the available context size"
 * - LM Studio: "greater than the context length"
 * - Kimi For Coding: "exceeded model token limit: X (requested: Y)"
-* - z.ai: "tokens in request more than max tokens allowed" or "Prompt exceeds max length"
 *
 * **Unreliable detection:**
 * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),
-*   sometimes returns rate limit errors instead of the explicit overflow error above. Pass
-*   contextWindow param to detect silent overflow.
+*   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.
 * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason "length" with
 *   output=0. Pass contextWindow param to detect via the "filled context + zero output" signal.
 * - Ollama: May truncate input silently for some setups, but may also return explicit
@@ -171,8 +160,7 @@ function resolveContextInputTokens(message) {
 */
 function isContextOverflow(message, contextWindow) {
 	if (message.stopReason === "error" && message.errorMessage) {
-		const errorMessage = message.errorMessage;
-		if (!NON_OVERFLOW_PATTERNS.some((p) => p.test(errorMessage)) && OVERFLOW_PATTERNS.some((p) => p.test(errorMessage))) return true;
+		if (!NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage)) && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage))) return true;
 	}
 	if (contextWindow && message.stopReason === "stop") {
 		const inputTokens = resolveContextInputTokens(message);
@@ -185,4 +173,4 @@ function isContextOverflow(message, contextWindow) {
 	return false;
 }
 //#endregion
-export { applyProviderReportedUsageCost, calculateCost, clampThinkingLevel, cleanupSessionResources, clearApiProviders, complete, completeSimple, createDeferredEventBuffer, createFirstStreamEventAbortController, createFirstStreamEventTimeoutError, createReasoningTagTextPartitioner, createSseByteGuard, decodeOpenAICodexJwtPayload, defaultApiRegistry, defaultLlmRuntime, findEnvKeys, getApiProvider, getApiProviders, getEnvApiKey, getFirstStreamEventTimeoutHandler, getFirstStreamEventTimeoutMs, getSupportedThinkingLevels, headersToRecord, isConfiguredContextSizeOverflowError, isContextOverflow, modelsAreEqual, notifyLlmRequestActivity, onLlmRequestActivity, parseJsonWithRepair, parseStreamingJson, registerSessionResourceCleanup, repairJson, resolveOpenAICodexAccountId, sanitizeSurrogates, shortHash, stream, streamSimple, withFirstStreamEventTimeout };
+export { applyProviderReportedUsageCost, calculateCost, clampThinkingLevel, cleanupSessionResources, clearApiProviders, complete, completeSimple, createDeferredEventBuffer, createFirstStreamEventAbortController, createFirstStreamEventTimeoutError, createReasoningTagTextPartitioner, createSseByteGuard, decodeOpenAICodexJwtPayload, defaultApiRegistry, defaultLlmRuntime, findEnvKeys, getApiProvider, getApiProviders, getEnvApiKey, getFirstStreamEventTimeoutHandler, getFirstStreamEventTimeoutMs, getSupportedThinkingLevels, headersToRecord, isConfiguredContextSizeOverflowError, isContextOverflow, modelsAreEqual, notifyLlmRequestActivity, onLlmRequestActivity, parseJsonWithRepair, parseStreamingJson, registerApiProvider, registerSessionResourceCleanup, repairJson, resolveOpenAICodexAccountId, sanitizeSurrogates, shortHash, stream, streamSimple, unregisterApiProviders, withFirstStreamEventTimeout };

package/dist/internal/shared.d.mts

@@ -1,6 +1,5 @@
-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-bzp5k29J.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-DRgdPqaZ.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-BZiVbGwK.mjs";
-import { n as normalizeStructuredPromptSection, r as sortPromptCacheToolsByName, t as normalizePromptCapabilityIds } from "../prompt-cache-stability-Cwcjv_fx.mjs";
 
 //#region packages/ai/src/providers/simple-options.d.ts
 type FirstEventStreamOptions = {
@@ -8,9 +7,6 @@ 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): {
   maxTokens: number;
@@ -18,20 +14,11 @@ declare function adjustMaxTokensForThinking(baseMaxTokens: number | undefined, m
 };
 //#endregion
 //#region packages/ai/src/providers/tool-result-text.d.ts
-/** Media metadata alone is not an attachment; provider emitters need inline bytes. */
-declare function hasMediaPayload(block: unknown): block is Record<string, unknown> & {
-  data: string;
-};
-/** Image metadata alone is not an attachment; provider emitters need inline bytes. */
-declare function isImageWithMediaPayload<T>(block: T): block is T & {
-  type: "image";
-  data: string;
-};
 declare function describeToolResultMediaPlaceholder(blocks: readonly unknown[]): string | undefined;
 declare function extractToolResultBlockText(block: unknown): string | undefined;
 declare function extractToolResultText(blocks: readonly unknown[]): string;
 //#endregion
-//#region packages/ai/src/transcript-transform.d.ts
+//#region packages/ai/src/providers/transform-messages.d.ts
 /**
  * Normalize tool call ID for cross-provider compatibility.
  * OpenAI Responses API generates IDs that are 450+ chars with special characters like `|`.
@@ -39,6 +26,12 @@ declare function extractToolResultText(blocks: readonly unknown[]): string;
  */
 declare function transformMessages<TApi extends Api>(messages: Message[], model: Model<TApi>, normalizeToolCallId?: (id: string, model: Model<TApi>, source: AssistantMessage) => string): Message[];
 //#endregion
+//#region packages/ai/src/utils/prompt-cache-stability.d.ts
+/** Normalize structured prompt text before hashing or snapshot comparison. */
+declare function normalizeStructuredPromptSection(text: string): string;
+/** Normalize, de-dupe, and sort capability ids for stable prompt payloads. */
+declare function normalizePromptCapabilityIds(capabilities: ReadonlyArray<string>): string[];
+//#endregion
 //#region packages/ai/src/utils/system-prompt-cache-boundary.d.ts
 declare const SYSTEM_PROMPT_CACHE_BOUNDARY = "\n<!-- OPENCLAW_CACHE_BOUNDARY -->\n";
 declare function stripSystemPromptCacheBoundary(text: string): string;
@@ -52,14 +45,4 @@ declare function prependSystemPromptAdditionAfterCacheBoundary(params: {
   systemPromptAddition?: string;
 }): string;
 //#endregion
-//#region packages/ai/src/utils/tls-certificate-errors.d.ts
-type TlsCertificateErrorKind = "certificate_invalid" | "hostname_mismatch";
-type TlsCertificateErrorDetails = {
-  kind: TlsCertificateErrorKind;
-  code?: string;
-  message: string;
-};
-/** Classify deterministic Node/OpenSSL certificate validation failures. */
-declare function inspectTlsCertificateError(error: unknown): TlsCertificateErrorDetails | null;
-//#endregion
-export { SYSTEM_PROMPT_CACHE_BOUNDARY, TlsCertificateErrorDetails, TlsCertificateErrorKind, adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToModel, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, inspectTlsCertificateError, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, sortPromptCacheToolsByName, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };
+export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };

package/dist/internal/shared.mjs

@@ -1,7 +1,3 @@
-import { i as transformMessages } from "../host-Dog2WQiR.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-DT5o51ur.mjs";
-import { a as isImageWithMediaPayload, c as ensureSystemPromptCacheBoundary, d as stripSystemPromptCacheBoundary, f as normalizePromptCapabilityIds, i as hasMediaPayload, l as prependSystemPromptAdditionAfterCacheBoundary, m as sortPromptCacheToolsByName, n as extractToolResultBlockText, p as normalizeStructuredPromptSection, r as extractToolResultText, s as SYSTEM_PROMPT_CACHE_BOUNDARY, t as describeToolResultMediaPlaceholder, u as splitSystemPromptCacheBoundary } from "../tool-result-text-CTpIRbYd.mjs";
-import { i as clampReasoning, n as buildBaseOptions, r as clampMaxTokensToModel, t as adjustMaxTokensForThinking } from "../simple-options-9lhRrN73.mjs";
-import "../transform-messages-C8mBqZxF.mjs";
-import { t as inspectTlsCertificateError } from "../tls-certificate-errors-DXSpluKI.mjs";
-export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToModel, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, hasMediaPayload, inspectTlsCertificateError, isImageWithMediaPayload, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, sortPromptCacheToolsByName, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };
+import { S as normalizeStructuredPromptSection, _ as ensureSystemPromptCacheBoundary, a as adjustMaxTokensForThinking, b as stripSystemPromptCacheBoundary, g as SYSTEM_PROMPT_CACHE_BOUNDARY, i as extractToolResultText, n as describeToolResultMediaPlaceholder, o as buildBaseOptions, r as extractToolResultBlockText, s as clampReasoning, t as transformMessages, v as prependSystemPromptAdditionAfterCacheBoundary, x as normalizePromptCapabilityIds, y as splitSystemPromptCacheBoundary } from "../transform-messages-DoWx35pj.mjs";
+export { SYSTEM_PROMPT_CACHE_BOUNDARY, adjustMaxTokensForThinking, buildBaseOptions, clampReasoning, describeToolResultMediaPlaceholder, ensureSystemPromptCacheBoundary, extractToolResultBlockText, extractToolResultText, normalizePromptCapabilityIds, normalizeStructuredPromptSection, prependSystemPromptAdditionAfterCacheBoundary, sanitizeSurrogates, splitSystemPromptCacheBoundary, stripSystemPromptCacheBoundary, transformMessages };

package/dist/{json-parse-BvXNt1-7.mjs → json-parse-DzNSIQBq.mjs}

@@ -42,7 +42,7 @@ function repairJson(json) {
 	let inString = false;
 	let stringValuePrefix = "";
 	for (let index = 0; index < json.length; index++) {
-		const char = json.charAt(index);
+		const char = json[index];
 		if (!inString) {
 			repaired += char;
 			if (char === "\"") {
@@ -58,8 +58,8 @@ function repairJson(json) {
 			continue;
 		}
 		if (char === "\\") {
-			const nextChar = json.charAt(index + 1);
-			if (!nextChar) {
+			const nextChar = json[index + 1];
+			if (nextChar === void 0) {
 				repaired += "\\\\";
 				continue;
 			}
@@ -96,7 +96,9 @@ function repairJson(json) {
 	return repaired;
 }
 function parseJsonWithRepair(json) {
-	return JSON.parse(repairJson(json));
+	const repairedJson = repairJson(json);
+	if (repairedJson !== json) return JSON.parse(repairedJson);
+	return JSON.parse(json);
 }
 function looksLikeWindowsPathPrefix(prefix) {
 	const tail = prefix.slice(-160);

package/dist/{llm-request-activity-BjtkplhG.mjs → llm-request-activity-CehVkZP-.mjs}

@@ -1,3 +1,21 @@
+//#region packages/ai/src/utils/deferred-event-buffer.ts
+function createDeferredEventBuffer(sink, onBufferedEvent) {
+	let events = [];
+	return {
+		push(event) {
+			events.push(event);
+			onBufferedEvent?.();
+		},
+		flush() {
+			for (const event of events) sink.push(event);
+			events = [];
+		},
+		discard() {
+			events = [];
+		}
+	};
+}
+//#endregion
 //#region packages/ai/src/utils/llm-request-activity.ts
 const requestActivityListeners = /* @__PURE__ */ new WeakMap();
 function notifyLlmRequestActivity(signal) {
@@ -14,4 +32,4 @@ function onLlmRequestActivity(signal, listener) {
 	};
 }
 //#endregion
-export { onLlmRequestActivity as n, notifyLlmRequestActivity as t };
+export { onLlmRequestActivity as n, createDeferredEventBuffer as r, notifyLlmRequestActivity as t };

package/dist/{mistral-CWmpvWYh.mjs → mistral-CBm3OQWD.mjs}

@@ -1,16 +1,12 @@
-import { n as getEnvApiKey } from "./env-api-keys-DrgeBuva.mjs";
-import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
-import { i as transformMessages, n as getAiTransportHost } from "./host-Dog2WQiR.mjs";
+import { n as getEnvApiKey } from "./env-api-keys-CtMlqaQ4.mjs";
+import { t as AssistantMessageEventStream } from "./event-stream-ReMmOTzX.mjs";
+import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Cn9KAz3z.mjs";
+import { n as parseStreamingJson } from "./json-parse-DzNSIQBq.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { a as isImageWithMediaPayload, d as stripSystemPromptCacheBoundary, o as truncateUtf16Safe, r as extractToolResultText, t as describeToolResultMediaPlaceholder } from "./tool-result-text-CTpIRbYd.mjs";
-import { c as calculateCost, l as clampThinkingLevel } from "./number-coercion-DvG7SNMg.mjs";
-import { n as parseStreamingJson } from "./json-parse-BvXNt1-7.mjs";
-import { d as transportAbortError } from "./transport-stream-shared-D81p90xq.mjs";
+import { b as stripSystemPromptCacheBoundary, i as extractToolResultText, n as describeToolResultMediaPlaceholder, o as buildBaseOptions, t as transformMessages } from "./transform-messages-DoWx35pj.mjs";
 import { t as shortHash } from "./hash-CHgqbJmD.mjs";
-import { n as buildBaseOptions, r as clampMaxTokensToModel } from "./simple-options-9lhRrN73.mjs";
-import "./transform-messages-C8mBqZxF.mjs";
 import { t as createSseByteGuard } from "./streaming-byte-guard-BrbkbwUu.mjs";
-import { randomUUID } from "node:crypto";
 import { HTTPClient, Mistral } from "@mistralai/mistralai";
 //#region packages/ai/src/providers/mistral.ts
 const MISTRAL_TOOL_CALL_ID_LENGTH = 9;
@@ -74,21 +70,13 @@ const streamMistral = (model, context, options) => {
 			let payload = buildChatPayload(model, context, transformMessages(context.messages, model, (id) => normalizeMistralToolCallId(id)), options);
 			const nextPayload = await options?.onPayload?.(payload, model);
 			if (nextPayload !== void 0) payload = nextPayload;
-			const headers = {
-				...model.headers,
-				...options?.headers
-			};
-			if (resolveMistralPromptCacheKey(options) && options?.sessionId) headers["x-affinity"] ||= options.sessionId;
-			const mistralStream = await mistral.chat.stream(payload, {
-				headers,
-				signal: options?.signal
-			});
+			const mistralStream = await mistral.chat.stream(payload, buildRequestOptions(model, options));
 			stream.push({
 				type: "start",
 				partial: output
 			});
 			await consumeChatStream(model, output, stream, mistralStream);
-			if (options?.signal?.aborted) throw transportAbortError(options.signal);
+			if (options?.signal?.aborted) throw new Error("Request was aborted");
 			if (output.stopReason === "aborted" || output.stopReason === "error") throw new Error("An unknown error occurred");
 			stream.push({
 				type: "done",
@@ -116,10 +104,7 @@ 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: ${model.provider}`);
-	const base = {
-		...buildBaseOptions(model, options, apiKey),
-		maxTokens: clampMaxTokensToModel(model, options?.maxTokens)
-	};
+	const base = buildBaseOptions(model, options, apiKey);
 	const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : void 0;
 	const reasoning = clampedReasoning === "off" ? void 0 : clampedReasoning;
 	const shouldUseReasoning = model.reasoning && reasoning !== void 0;
@@ -177,7 +162,7 @@ function deriveMistralToolCallId(id, attempt) {
 	const normalized = id.replace(/[^a-zA-Z0-9]/g, "");
 	if (attempt === 0 && normalized.length === MISTRAL_TOOL_CALL_ID_LENGTH) return normalized;
 	const seedBase = normalized || id;
-	return shortHash(attempt === 0 ? seedBase : `${seedBase}:${attempt}`).replace(/[^a-zA-Z0-9]/g, "").padEnd(MISTRAL_TOOL_CALL_ID_LENGTH, "0").slice(0, MISTRAL_TOOL_CALL_ID_LENGTH);
+	return shortHash(attempt === 0 ? seedBase : `${seedBase}:${attempt}`).replace(/[^a-zA-Z0-9]/g, "").slice(0, MISTRAL_TOOL_CALL_ID_LENGTH);
 }
 function formatMistralError(error) {
 	if (error instanceof Error) {
@@ -192,8 +177,7 @@ function formatMistralError(error) {
 }
 function truncateErrorText(text, maxChars) {
 	if (text.length <= maxChars) return text;
-	const truncated = truncateUtf16Safe(text, maxChars);
-	return `${truncated}... [truncated ${text.length - truncated.length} chars]`;
+	return `${text.slice(0, maxChars)}... [truncated ${text.length - maxChars} chars]`;
 }
 function safeJsonStringify(value) {
 	try {
@@ -203,6 +187,16 @@ function safeJsonStringify(value) {
 		return String(value);
 	}
 }
+function buildRequestOptions(model, options) {
+	const requestOptions = { retries: { strategy: "none" } };
+	if (options?.signal) requestOptions.signal = options.signal;
+	const headers = {};
+	if (model.headers) Object.assign(headers, model.headers);
+	if (options?.headers) Object.assign(headers, options.headers);
+	if (options?.sessionId && !headers["x-affinity"]) headers["x-affinity"] = options.sessionId;
+	if (Object.keys(headers).length > 0) requestOptions.headers = headers;
+	return requestOptions;
+}
 function buildChatPayload(model, context, messages, options) {
 	const payload = {
 		model: model.id,
@@ -224,86 +218,17 @@ 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;
 	const blockIndex = () => blocks.length - 1;
-	const toolBlockIdentities = /* @__PURE__ */ new Map();
-	const normalizeMissingToolCallId = createMistralToolCallIdNormalizer();
-	const missingToolCallIdScope = randomUUID();
-	const createMissingToolCallId = (contentIndex) => normalizeMissingToolCallId(`${missingToolCallIdScope}:toolcall:${contentIndex}`);
-	const findIdentityCandidates = (matches, excludedContentIndexes) => {
-		const candidates = /* @__PURE__ */ new Set();
-		for (const [contentIndex, identity] of toolBlockIdentities) if (!excludedContentIndexes?.has(contentIndex) && matches(identity)) candidates.add(contentIndex);
-		return candidates;
-	};
-	const intersectCandidates = (left, right) => new Set([...left].filter((contentIndex) => right.has(contentIndex)));
-	const requireSingleCandidate = (candidates) => {
-		if (candidates.size > 1) throw new Error("Mistral streamed tool-call continuation is ambiguous; refusing to merge arguments");
-		return candidates.values().next().value;
-	};
-	const requireExistingCandidate = (candidates) => {
-		const candidate = requireSingleCandidate(candidates);
-		if (candidate === void 0) throw new Error("Mistral streamed tool-call identities conflict; refusing to merge arguments");
-		return candidate;
-	};
-	const resolveToolBlockIndex = (params) => {
-		const explicitId = params.explicitId;
-		const functionName = params.functionName;
-		const toolCallIndex = params.index;
-		const idCandidates = explicitId ? findIdentityCandidates((identity) => identity.explicitIds.has(explicitId), params.usedContentIndexes) : /* @__PURE__ */ new Set();
-		const nameCandidates = functionName ? findIdentityCandidates((identity) => identity.functionNames.has(functionName), params.usedContentIndexes) : /* @__PURE__ */ new Set();
-		const indexCandidates = toolCallIndex === void 0 ? /* @__PURE__ */ new Set() : findIdentityCandidates((identity) => identity.indexes.has(toolCallIndex), params.usedContentIndexes);
-		if (idCandidates.size > 0) {
-			let candidates = idCandidates;
-			if (nameCandidates.size > 0) candidates = intersectCandidates(candidates, nameCandidates);
-			return requireExistingCandidate(candidates);
-		}
-		if (nameCandidates.size > 0) {
-			const idCompatibleCandidates = new Set([...nameCandidates].filter((contentIndex) => {
-				const identity = toolBlockIdentities.get(contentIndex);
-				if (!identity) return false;
-				return !explicitId || identity.explicitIds.size === 0;
-			}));
-			if (idCompatibleCandidates.size <= 1 && (toolCallIndex === void 0 || toolCallIndex === 0)) return requireSingleCandidate(idCompatibleCandidates);
-			const indexCompatibleCandidates = new Set([...idCompatibleCandidates].filter((contentIndex) => {
-				const identity = toolBlockIdentities.get(contentIndex);
-				if (!identity) return false;
-				return toolCallIndex === void 0 || identity.indexes.size === 0 || identity.indexes.has(toolCallIndex);
-			}));
-			if (indexCompatibleCandidates.size === 0) return;
-			return requireSingleCandidate(indexCompatibleCandidates);
-		}
-		if (functionName) {
-			const namelessCandidates = new Set([...indexCandidates].filter((contentIndex) => {
-				const identity = toolBlockIdentities.get(contentIndex);
-				return identity?.functionNames.size === 0 && (!explicitId || identity.explicitIds.size === 0);
-			}));
-			return requireSingleCandidate(namelessCandidates);
-		}
-		if (explicitId) {
-			const idlessCandidates = new Set([...indexCandidates].filter((contentIndex) => toolBlockIdentities.get(contentIndex)?.explicitIds.size === 0));
-			return requireSingleCandidate(idlessCandidates);
-		}
-		return requireSingleCandidate(indexCandidates);
-	};
+	const toolBlocksByKey = /* @__PURE__ */ new Map();
 	const finishCurrentBlock = (block) => {
 		if (!block) return;
 		if (block.type === "text") {
@@ -326,13 +251,11 @@ async function consumeChatStream(model, output, stream, mistralStream) {
 		const chunk = event.data;
 		output.responseId ||= chunk.id;
 		if (chunk.usage) {
-			const promptTokens = chunk.usage.promptTokens || 0;
-			const cachedPromptTokens = readMistralCachedPromptTokens(chunk.usage, promptTokens);
-			output.usage.input = Math.max(0, promptTokens - cachedPromptTokens);
+			output.usage.input = chunk.usage.promptTokens || 0;
 			output.usage.output = chunk.usage.completionTokens || 0;
-			output.usage.cacheRead = cachedPromptTokens;
+			output.usage.cacheRead = 0;
 			output.usage.cacheWrite = 0;
-			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output + output.usage.cacheRead;
+			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;
 			calculateCost(model, output.usage);
 		}
 		const choice = chunk.choices[0];
@@ -417,77 +340,52 @@ async function consumeChatStream(model, output, stream, mistralStream) {
 			}
 		}
 		const toolCalls = delta.toolCalls || [];
-		const usedToolBlockIndexes = /* @__PURE__ */ new Set();
 		for (const toolCall of toolCalls) {
 			if (currentBlock) {
 				finishCurrentBlock(currentBlock);
 				currentBlock = null;
 			}
-			const toolCallIndex = typeof toolCall.index === "number" && Number.isInteger(toolCall.index) ? toolCall.index : void 0;
-			const providedCallId = toolCall.id && toolCall.id !== "null" ? toolCall.id : void 0;
-			const functionName = toolCall.function.name.trim() || void 0;
-			const existingIndex = resolveToolBlockIndex({
-				explicitId: providedCallId,
-				functionName,
-				index: toolCallIndex,
-				usedContentIndexes: usedToolBlockIndexes
-			});
+			const callId = toolCall.id && toolCall.id !== "null" ? toolCall.id : deriveMistralToolCallId(`toolcall:${toolCall.index ?? 0}`, 0);
+			const key = `${callId}:${toolCall.index || 0}`;
+			const existingIndex = toolBlocksByKey.get(key);
 			let block;
 			if (existingIndex !== void 0) {
 				const existing = output.content[existingIndex];
 				if (existing?.type === "toolCall") block = existing;
 			}
 			if (!block) {
-				const contentIndex = output.content.length;
 				block = {
 					type: "toolCall",
-					id: providedCallId ?? createMissingToolCallId(contentIndex),
-					name: functionName ?? "",
+					id: callId,
+					name: toolCall.function.name,
 					arguments: {},
 					partialArgs: ""
 				};
 				output.content.push(block);
-				toolBlockIdentities.set(contentIndex, {
-					explicitIds: new Set(providedCallId ? [providedCallId] : []),
-					functionNames: new Set(functionName ? [functionName] : []),
-					indexes: new Set(toolCallIndex === void 0 ? [] : [toolCallIndex])
-				});
+				toolBlocksByKey.set(key, output.content.length - 1);
 				stream.push({
 					type: "toolcall_start",
-					contentIndex,
+					contentIndex: output.content.length - 1,
 					partial: output
 				});
 			}
-			const contentIndex = output.content.indexOf(block);
-			const identity = toolBlockIdentities.get(contentIndex);
-			if (!identity) throw new Error("Mistral streamed tool-call identity is missing");
-			usedToolBlockIndexes.add(contentIndex);
-			if (providedCallId) {
-				block.id = providedCallId;
-				identity.explicitIds.add(providedCallId);
-			}
-			if (functionName) {
-				if (identity.functionNames.size > 0 && !identity.functionNames.has(functionName)) throw new Error("Mistral streamed tool-call continuation changed function name; refusing to merge arguments");
-				block.name = functionName;
-				identity.functionNames.add(functionName);
-			}
-			if (toolCallIndex !== void 0) identity.indexes.add(toolCallIndex);
 			const argsDelta = typeof toolCall.function.arguments === "string" ? toolCall.function.arguments : JSON.stringify(toolCall.function.arguments || {});
 			block.partialArgs = (block.partialArgs || "") + argsDelta;
 			block.arguments = parseStreamingJson(block.partialArgs);
 			stream.push({
 				type: "toolcall_delta",
-				contentIndex,
+				contentIndex: toolBlocksByKey.get(key),
 				delta: argsDelta,
 				partial: output
 			});
 		}
 	}
 	finishCurrentBlock(currentBlock);
-	for (const index of toolBlockIdentities.keys()) {
-		const block = output.content.at(index);
-		if (block?.type !== "toolCall") continue;
+	for (const index of toolBlocksByKey.values()) {
+		const block = output.content[index];
+		if (block.type !== "toolCall") continue;
 		const toolBlock = block;
+		toolBlock.arguments = parseStreamingJson(toolBlock.partialArgs);
 		delete toolBlock.partialArgs;
 		stream.push({
 			type: "toolcall_end",
@@ -595,14 +493,14 @@ function toChatMessages(messages, supportsImages) {
 			continue;
 		}
 		const toolContent = [];
-		const toolText = buildToolResultText(extractToolResultText(msg.content), describeToolResultMediaPlaceholder(msg.content), msg.content.some(isImageWithMediaPayload), supportsImages, msg.isError);
+		const toolText = buildToolResultText(extractToolResultText(msg.content), describeToolResultMediaPlaceholder(msg.content), msg.content.some((part) => part.type === "image"), supportsImages, msg.isError);
 		toolContent.push({
 			type: "text",
 			text: toolText
 		});
 		for (const part of msg.content) {
 			if (!supportsImages) continue;
-			if (!isImageWithMediaPayload(part)) continue;
+			if (part.type !== "image") continue;
 			toolContent.push({
 				type: "image_url",
 				imageUrl: `data:${part.mimeType};base64,${part.data}`

package/dist/model-utils-Cn9KAz3z.mjs

@@ -0,0 +1,69 @@
+import { a as requiresClaudeMandatoryAdaptiveThinking, l as resolveClaudeNativeThinkingLevelMap } from "./src-CjOfhrH3.mjs";
+//#region packages/ai/src/model-utils.ts
+/** Calculates and stores model cost fields from token usage and per-million pricing. */
+function calculateCost(model, usage) {
+	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.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;
+	return usage.cost;
+}
+/** Replaces the catalog estimate when the provider reports an authoritative billed total. */
+function applyProviderReportedUsageCost(usage, reportedCost) {
+	if (typeof reportedCost !== "number" || !Number.isFinite(reportedCost) || reportedCost < 0) return;
+	usage.cost.total = reportedCost;
+	usage.cost.totalOrigin = "provider-billed";
+}
+const EXTENDED_THINKING_LEVELS = [
+	"off",
+	"minimal",
+	"low",
+	"medium",
+	"high",
+	"xhigh",
+	"max"
+];
+function resolveThinkingLevelMap(model) {
+	return model.api === "anthropic-messages" ? resolveClaudeNativeThinkingLevelMap(model) ?? model.thinkingLevelMap : model.thinkingLevelMap;
+}
+/** Returns thinking levels exposed by a reasoning-capable model. */
+function getSupportedThinkingLevels(model) {
+	const mandatoryAdaptiveContract = model.api === "anthropic-messages" && requiresClaudeMandatoryAdaptiveThinking(model);
+	if (!model.reasoning && !mandatoryAdaptiveContract) return ["off"];
+	const thinkingLevelMap = resolveThinkingLevelMap(model);
+	return EXTENDED_THINKING_LEVELS.filter((level) => {
+		const mapped = thinkingLevelMap?.[level];
+		if (mapped === null) return false;
+		if (level === "xhigh" || level === "max") return mapped !== void 0;
+		return true;
+	});
+}
+/** Clamps a requested thinking level to the closest supported level for a model. */
+function clampThinkingLevel(model, level) {
+	const availableLevels = getSupportedThinkingLevels(model);
+	if (availableLevels.includes(level)) return level;
+	const requestedIndex = EXTENDED_THINKING_LEVELS.indexOf(level);
+	if (requestedIndex === -1) return availableLevels[0] ?? "off";
+	const thinkingLevelMap = resolveThinkingLevelMap(model);
+	if ((level === "xhigh" || level === "max") && thinkingLevelMap?.[level] === null) for (let i = requestedIndex - 1; i >= 0; i--) {
+		const candidate = EXTENDED_THINKING_LEVELS[i];
+		if (availableLevels.includes(candidate)) return candidate;
+	}
+	for (let i = requestedIndex; i < EXTENDED_THINKING_LEVELS.length; i++) {
+		const candidate = EXTENDED_THINKING_LEVELS[i];
+		if (availableLevels.includes(candidate)) return candidate;
+	}
+	for (let i = requestedIndex - 1; i >= 0; i--) {
+		const candidate = EXTENDED_THINKING_LEVELS[i];
+		if (availableLevels.includes(candidate)) return candidate;
+	}
+	return availableLevels[0] ?? "off";
+}
+/** Compares model identity by provider and id. */
+function modelsAreEqual(a, b) {
+	if (!a || !b) return false;
+	return a.id === b.id && a.provider === b.provider;
+}
+//#endregion
+export { modelsAreEqual as a, getSupportedThinkingLevels as i, calculateCost as n, clampThinkingLevel as r, applyProviderReportedUsageCost as t };