@openclaw/ai 2026.7.2-beta.2 → 2026.7.2-beta.4

package/dist/internal/openai.d.mts

@@ -1,279 +1,2 @@
-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 OpenAI from "openai";
-import { TSchema } from "typebox";
-import { ChatCompletionMessageParam } from "openai/resources/chat/completions.js";
-import { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
-
-//#region packages/ai/src/providers/agent-tools-parameter-schema.d.ts
-/**
- * Narrow structural view of the host's model compat config. packages/ai must stay
- * config-agnostic, so only tool-schema-relevant fields are modeled here; the host's
- * ModelCompatConfig remains structurally assignable.
- */
-type ToolSchemaModelCompat = {
-  toolSchemaProfile?: string;
-  unsupportedToolSchemaKeywords?: string[];
-  omitEmptyArrayItems?: boolean;
-};
-/** Extracts the compat record whether callers pass a model (`{ compat }`) or the compat itself. */
-declare function extractToolSchemaModelCompat(modelOrCompat: {
-  compat?: unknown;
-} | ToolSchemaModelCompat | undefined): ToolSchemaModelCompat | undefined;
-/** JSON Schema keywords this model/provider rejects in tool schemas. */
-declare function resolveUnsupportedToolSchemaKeywords(modelOrCompat: {
-  compat?: unknown;
-} | ToolSchemaModelCompat | undefined): ReadonlySet<string>;
-/** Whether empty `items: {}` on array schemas must be omitted for this model/provider. */
-declare function shouldOmitEmptyArrayItems(modelOrCompat: {
-  compat?: unknown;
-} | ToolSchemaModelCompat | undefined): boolean;
-type ToolParameterSchemaOptions = {
-  modelProvider?: string;
-  modelId?: string;
-  modelCompat?: ToolSchemaModelCompat;
-};
-/** Return a provider-compatible JSON schema for a model-facing tool. */
-declare function normalizeToolParameterSchema(schema: unknown, options?: ToolParameterSchemaOptions): TSchema;
-//#endregion
-//#region packages/ai/src/providers/azure-deployment-map.d.ts
-/** Parses AZURE_OPENAI_DEPLOYMENT_MAP-style model=deployment entries. */
-declare function parseAzureDeploymentNameMap(value: string | undefined): Map<string, string>;
-/**
- * Resolves the Azure deployment name for a model id, falling back to the model id.
- *
- * An exact-case match always wins, so configs that intentionally distinguish keys by
- * case keep their exact mappings; a case-insensitive match is only used as a fallback
- * (e.g. `GPT-4o` against a `gpt-4o=...` map) to avoid 404s from casing differences.
- */
-declare function resolveAzureDeploymentNameFromMap(params: {
-  modelId: string;
-  deploymentMap?: string;
-}): string;
-//#endregion
-//#region packages/ai/src/providers/azure-openai-responses-client-compat.d.ts
-declare function isTraditionalAzureOpenAIHost(hostname: string): boolean;
-declare function isOpenAICompatibleAzureResponsesBaseUrl(baseUrl: string): boolean;
-//#endregion
-//#region packages/ai/src/providers/clean-for-gemini.d.ts
-declare const GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS: Set<string>;
-declare function cleanSchemaForGemini(schema: unknown): TSchema;
-//#endregion
-//#region packages/ai/src/providers/openai-tool-projection.d.ts
-type OpenAIToolDescriptor = {
-  readonly name?: unknown;
-  readonly description?: unknown;
-  readonly parameters: unknown;
-};
-type OpenAIProjectedTool = {
-  readonly toolIndex: number;
-  readonly name: string;
-  readonly description?: string;
-  readonly parameters: Record<string, unknown>;
-};
-type OpenAIToolProjectionDiagnostic = {
-  readonly toolIndex: number;
-  readonly toolName?: string;
-  readonly violations: readonly string[];
-};
-type OpenAIToolProjection = {
-  readonly inputToolCount: number;
-  readonly tools: readonly OpenAIProjectedTool[];
-  readonly diagnostics: readonly OpenAIToolProjectionDiagnostic[];
-};
-type OpenAIResponsesToolChoice = ResponseCreateParamsStreaming["tool_choice"];
-type OpenAICompletionsSdkToolChoice = OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming["tool_choice"];
-type OpenAICompletionsToolChoice = Exclude<OpenAICompletionsSdkToolChoice, {
-  type: "custom";
-}>;
-/** Snapshots direct/custom tool descriptors before OpenAI payload construction. */
-declare function projectOpenAITools(tools: readonly OpenAIToolDescriptor[]): OpenAIToolProjection;
-/** Keeps Responses tool choices aligned with surviving function schemas. */
-declare function reconcileOpenAIResponsesToolChoice(choice: OpenAIResponsesToolChoice, projection: OpenAIToolProjection): OpenAIResponsesToolChoice | undefined;
-/** Keeps Chat Completions tool choices aligned with surviving function schemas. */
-declare function reconcileOpenAICompletionsToolChoice(choice: OpenAICompletionsSdkToolChoice, projection: OpenAIToolProjection): OpenAICompletionsSdkToolChoice | undefined;
-//#endregion
-//#region packages/ai/src/providers/openai-completions.d.ts
-interface OpenAICompletionsOptions extends StreamOptions {
-  toolChoice?: OpenAICompletionsToolChoice;
-  reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
-}
-type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "cacheControlFormat"> & {
-  cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];
-};
-declare const streamOpenAICompletions: StreamFunction<"openai-completions", OpenAICompletionsOptions>;
-declare const streamSimpleOpenAICompletions: StreamFunction<"openai-completions", SimpleStreamOptions>;
-declare function convertMessages(model: Model<"openai-completions">, context: Context, compat: ResolvedOpenAICompletionsCompat, options?: {
-  cacheOptOutIndexes?: Set<number>;
-  preserveSystemPromptCacheBoundary?: boolean;
-}): ChatCompletionMessageParam[];
-//#endregion
-//#region packages/ai/src/providers/openai-prompt-cache.d.ts
-/** Maximum prompt cache key length accepted by OpenAI-compatible request metadata. */
-declare const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;
-/** Truncates a prompt cache key by Unicode code point count. */
-declare function clampOpenAIPromptCacheKey(key: string | undefined): string | undefined;
-//#endregion
-//#region packages/ai/src/providers/openai-reasoning-effort.d.ts
-type OpenAIReasoningEffort = "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
-type OpenAIApiReasoningEffort = OpenAIReasoningEffort | (string & {});
-type OpenAIReasoningModel = {
-  provider?: unknown;
-  id?: unknown;
-  name?: unknown;
-  api?: unknown;
-  baseUrl?: unknown;
-  compat?: unknown;
-};
-/** Return whether a model is the GPT-5.4 mini family. */
-declare function isOpenAIGpt54MiniModel(model: OpenAIReasoningModel): boolean;
-/** Return whether a model is the GPT-5.5 family. */
-declare function isOpenAIGpt55Model(model: OpenAIReasoningModel): boolean;
-/** Return whether a model is the GPT-5.6 family. */
-declare function isOpenAIGpt56Model(model: OpenAIReasoningModel): boolean;
-/** Normalize user-facing reasoning effort names to API effort names. */
-declare function normalizeOpenAIReasoningEffort(effort: string): string;
-/** Resolve the reasoning efforts accepted by a specific OpenAI-compatible model. */
-declare function resolveOpenAISupportedReasoningEfforts(model: OpenAIReasoningModel): readonly OpenAIApiReasoningEffort[];
-/**
- * Return whether a model accepts the temperature parameter. The GPT-5.6
- * family rejects it with a 400; catalog compat can override per model.
- */
-declare function supportsOpenAITemperature(model: OpenAIReasoningModel): boolean;
-/** Return whether a model accepts a requested reasoning effort. */
-declare function supportsOpenAIReasoningEffort(model: OpenAIReasoningModel, effort: string): boolean;
-/** Resolve a requested reasoning effort to the closest value supported by the model. */
-declare function resolveOpenAIReasoningEffortForModel(params: {
-  model: OpenAIReasoningModel;
-  effort: string;
-  fallbackMap?: Record<string, string>;
-}): OpenAIApiReasoningEffort | undefined;
-//#endregion
-//#region packages/ai/src/providers/openai-responses.d.ts
-interface OpenAIResponsesOptions extends StreamOptions {
-  reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
-  reasoningSummary?: "auto" | "detailed" | "concise" | null;
-  replayResponsesItemIds?: boolean;
-  serviceTier?: ResponseCreateParamsStreaming["service_tier"];
-}
-/**
- * Generate function for OpenAI Responses API
- */
-declare const streamOpenAIResponses: StreamFunction<"openai-responses", OpenAIResponsesOptions>;
-declare const streamSimpleOpenAIResponses: StreamFunction<"openai-responses", SimpleStreamOptions>;
-//#endregion
-//#region packages/ai/src/providers/openai-responses-stream-compat.d.ts
-declare const OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE = "output_text";
-declare const AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE = "text";
-declare const OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE = "response.output_text.delta";
-declare const AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE = "response.text.delta";
-type ResponsesTextContentPartType = typeof OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE | typeof AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE;
-type ResponsesTextDeltaEventType = typeof OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE | typeof AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE;
-type AzureResponsesTextContentPart = {
-  type: typeof AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE;
-  text: string;
-};
-type AzureResponsesTextDeltaEvent = {
-  type: typeof AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE;
-  delta: string;
-};
-declare function isResponsesTextContentPartType(type: unknown): type is ResponsesTextContentPartType;
-declare function isResponsesTextDeltaEventType(type: unknown): type is ResponsesTextDeltaEventType;
-declare function isAzureResponsesTextDeltaEventType(type: unknown): type is typeof AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE;
-declare function isAzureResponsesTextDeltaEvent(event: {
-  type?: unknown;
-  delta?: unknown;
-}): event is AzureResponsesTextDeltaEvent;
-type ResponsesMessageSnapshotCollapse = {
-  kind: "extend";
-  text: string;
-} | {
-  kind: "keep";
-};
-declare function resolveResponsesMessageSnapshotCollapse(params: {
-  prior: {
-    text: string;
-    phase: string | undefined;
-  } | null;
-  nextText: string;
-  nextPhase: string | undefined;
-}): ResponsesMessageSnapshotCollapse;
-//#endregion
-//#region packages/ai/src/providers/openai-responses-tool-call-tracker.d.ts
-type ResponsesToolCallIdentity = {
-  itemId?: string;
-  callId?: string;
-};
-type ResponsesToolCallState = ResponsesToolCallIdentity & {
-  argumentStreamReliable: boolean;
-};
-type ResponsesToolCallEvent = {
-  output_index?: unknown;
-  item_id?: unknown;
-};
-declare function readResponsesToolCallItemIdentity(item: {
-  id?: unknown;
-  call_id?: unknown;
-}): ResponsesToolCallIdentity;
-declare function createResponsesToolCallTracker<TState extends ResponsesToolCallState>(): {
-  register(event: ResponsesToolCallEvent, state: TState): void;
-  resolve(event: ResponsesToolCallEvent, identity?: ResponsesToolCallIdentity): TState | undefined;
-  forget(toolCall: TState): void;
-  markArgumentsUnreliable(): void;
-  hasActive(): boolean;
-};
-//#endregion
-//#region packages/ai/src/providers/openai-stop-reason.d.ts
-type OpenAIStopReasonResult = {
-  stopReason: StopReason;
-  errorMessage?: string;
-};
-declare function mapOpenAIStopReason(reason: string | null, options?: {
-  allowSingularToolCall?: boolean;
-}): OpenAIStopReasonResult;
-//#endregion
-//#region packages/ai/src/providers/openai-tool-schema.d.ts
-/**
- * OpenAI strict-tool-schema normalization and diagnostics.
- *
- * Strict schemas need all object properties required and `additionalProperties: false`; model
- * compatibility settings can also remove unsupported schema constructs before strict checks run.
- */
-type ToolSchemaCompatInput = {
-  unsupportedToolSchemaKeywords?: unknown;
-  omitEmptyArrayItems?: unknown;
-};
-declare function clearOpenAIToolSchemaCacheForTest(): void;
-/** Normalizes a tool parameter schema into the OpenAI strict JSON-schema subset. */
-declare function normalizeStrictOpenAIJsonSchema(schema: unknown, modelCompat?: ToolSchemaCompatInput | null): unknown;
-/** Normalizes tool parameters using strict OpenAI rules only when strict mode is active. */
-declare function normalizeOpenAIStrictToolParameters<T>(schema: T, strict: boolean, modelCompat?: ToolSchemaCompatInput | null): T;
-/** Returns whether a schema already satisfies OpenAI strict tool-schema constraints. */
-declare function isStrictOpenAIJsonSchemaCompatible(schema: unknown): boolean;
-type OpenAIStrictToolSchemaDiagnostic = {
-  toolIndex: number;
-  toolName?: string;
-  violations: string[];
-};
-/** Returns strict-schema diagnostics for an already materialized OpenAI tool projection. */
-declare function findOpenAIStrictToolProjectionDiagnostics(projection: OpenAIToolProjection): OpenAIStrictToolSchemaDiagnostic[];
-/** Resolves strict mode for the projected tools that will be emitted in the request payload. */
-declare function resolveOpenAIProjectedToolsStrictToolFlag(projection: OpenAIToolProjection, strict: boolean | null | undefined): boolean | undefined;
-//#endregion
-//#region packages/ai/src/providers/schema-keyword-strip.d.ts
-/** Recursively remove schema keywords unsupported by a target provider/tool surface. */
-declare function stripUnsupportedSchemaKeywords(schema: unknown, unsupportedKeywords: ReadonlySet<string>): unknown;
-//#endregion
-//#region packages/ai/src/providers/tool-schema-json-projection.d.ts
-/** JSON-safe schema value used when projecting runtime tool parameters. */
-type RuntimeToolInputSchemaJson = null | boolean | number | string | RuntimeToolInputSchemaJson[] | {
-  [key: string]: RuntimeToolInputSchemaJson;
-};
-/** Projected runtime tool schema plus validation violations. */
-type RuntimeToolInputSchemaProjection = {
-  readonly schema: RuntimeToolInputSchemaJson;
-  readonly violations: readonly string[];
-};
-/** Projects one runtime tool input schema to JSON and reports runtime incompatibilities. */
-declare function projectRuntimeToolInputSchema(schema: unknown, path?: string): RuntimeToolInputSchemaProjection;
-//#endregion
-export { AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, AzureResponsesTextContentPart, AzureResponsesTextDeltaEvent, GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, OpenAIApiReasoningEffort, OpenAICompletionsOptions, OpenAICompletionsToolChoice, OpenAIReasoningEffort, OpenAIResponsesOptions, OpenAIStopReasonResult, OpenAIToolProjection, ResponsesMessageSnapshotCollapse, ResponsesTextContentPartType, ResponsesTextDeltaEventType, ResponsesToolCallIdentity, ResponsesToolCallState, RuntimeToolInputSchemaJson, RuntimeToolInputSchemaProjection, ToolParameterSchemaOptions, ToolSchemaModelCompat, 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 };
+import { $ as convertMessages, A as ResponsesTextContentPartType, B as OpenAIApiReasoningEffort, C as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, D as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, E as AzureResponsesTextDeltaEvent, F as isResponsesTextDeltaEventType, G as normalizeOpenAIReasoningEffort, H as isOpenAIGpt54MiniModel, I as resolveResponsesMessageSnapshotCollapse, J as supportsOpenAIReasoningEffort, K as resolveOpenAIReasoningEffortForModel, L as OpenAIResponsesOptions, M as isAzureResponsesTextDeltaEvent, N as isAzureResponsesTextDeltaEventType, O as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, P as isResponsesTextContentPartType, Q as OpenAICompletionsOptions, R as streamOpenAIResponses, S as resolveResponsesTerminalStopReason, T as AzureResponsesTextContentPart, U as isOpenAIGpt55Model, V as OpenAIReasoningEffort, W as isOpenAIGpt56Model, X as OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, Y as supportsOpenAITemperature, Z as clampOpenAIPromptCacheKey, _ as createResponsesToolCallTracker, _t as resolveUnsupportedToolSchemaKeywords, a as clearOpenAIToolSchemaCacheForTest, at as reconcileOpenAICompletionsToolChoice, b as mapResponsesTerminalUsage, c as normalizeOpenAIStrictToolParameters, ct as cleanSchemaForGemini, d as findOpenAIStrictSchemaViolations, dt as parseAzureDeploymentNameMap, et as streamOpenAICompletions, f as normalizeOpenAIStrictCompatSchema, ft as resolveAzureDeploymentNameFromMap, g as ResponsesToolCallState, gt as normalizeToolParameterSchema, h as ResponsesToolCallIdentity, ht as extractToolSchemaModelCompat, i as stripUnsupportedSchemaKeywords, it as projectOpenAITools, j as ResponsesTextDeltaEventType, k as ResponsesMessageSnapshotCollapse, l as normalizeStrictOpenAIJsonSchema, lt as isOpenAICompatibleAzureResponsesBaseUrl, m as mapOpenAIStopReason, mt as ToolSchemaModelCompat, n as RuntimeToolInputSchemaProjection, nt as OpenAICompletionsToolChoice, o as findOpenAIStrictToolProjectionDiagnostics, ot as reconcileOpenAIResponsesToolChoice, p as OpenAIStopReasonResult, pt as ToolParameterSchemaOptions, q as resolveOpenAISupportedReasoningEfforts, r as projectRuntimeToolInputSchema, rt as OpenAIToolProjection, s as isStrictOpenAIJsonSchemaCompatible, st as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, t as RuntimeToolInputSchemaJson, tt as streamSimpleOpenAICompletions, u as resolveOpenAIProjectedToolsStrictToolFlag, ut as isTraditionalAzureOpenAIHost, v as readResponsesToolCallItemIdentity, vt as shouldOmitEmptyArrayItems, w as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, x as readResponsesReasoningTokens, y as ResponsesTerminalUsagePayload, z as streamSimpleOpenAIResponses } from "../openai-CoGicoDt.mjs";
+export { AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, AzureResponsesTextContentPart, AzureResponsesTextDeltaEvent, GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, OpenAIApiReasoningEffort, OpenAICompletionsOptions, OpenAICompletionsToolChoice, OpenAIReasoningEffort, OpenAIResponsesOptions, OpenAIStopReasonResult, OpenAIToolProjection, ResponsesMessageSnapshotCollapse, ResponsesTerminalUsagePayload, ResponsesTextContentPartType, ResponsesTextDeltaEventType, ResponsesToolCallIdentity, ResponsesToolCallState, RuntimeToolInputSchemaJson, RuntimeToolInputSchemaProjection, ToolParameterSchemaOptions, ToolSchemaModelCompat, clampOpenAIPromptCacheKey, cleanSchemaForGemini, clearOpenAIToolSchemaCacheForTest, convertMessages, createResponsesToolCallTracker, extractToolSchemaModelCompat, findOpenAIStrictSchemaViolations, findOpenAIStrictToolProjectionDiagnostics, isAzureResponsesTextDeltaEvent, isAzureResponsesTextDeltaEventType, isOpenAICompatibleAzureResponsesBaseUrl, isOpenAIGpt54MiniModel, isOpenAIGpt55Model, isOpenAIGpt56Model, isResponsesTextContentPartType, isResponsesTextDeltaEventType, isStrictOpenAIJsonSchemaCompatible, isTraditionalAzureOpenAIHost, mapOpenAIStopReason, mapResponsesTerminalUsage, normalizeOpenAIReasoningEffort, normalizeOpenAIStrictCompatSchema, normalizeOpenAIStrictToolParameters, normalizeStrictOpenAIJsonSchema, normalizeToolParameterSchema, parseAzureDeploymentNameMap, projectOpenAITools, projectRuntimeToolInputSchema, readResponsesReasoningTokens, readResponsesToolCallItemIdentity, reconcileOpenAICompletionsToolChoice, reconcileOpenAIResponsesToolChoice, resolveAzureDeploymentNameFromMap, resolveOpenAIProjectedToolsStrictToolFlag, resolveOpenAIReasoningEffortForModel, resolveOpenAISupportedReasoningEfforts, resolveResponsesMessageSnapshotCollapse, resolveResponsesTerminalStopReason, resolveUnsupportedToolSchemaKeywords, shouldOmitEmptyArrayItems, streamOpenAICompletions, streamOpenAIResponses, streamSimpleOpenAICompletions, streamSimpleOpenAIResponses, stripUnsupportedSchemaKeywords, supportsOpenAIReasoningEffort, supportsOpenAITemperature };

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 mapOpenAIStopReason, i as streamSimpleOpenAICompletions, r as streamOpenAICompletions, t as convertMessages } from "../openai-completions-exO8NFQf.mjs";
+import { A as isOpenAIGpt54MiniModel, B as resolveUnsupportedToolSchemaKeywords, C as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, D as isResponsesTextContentPartType, E as isAzureResponsesTextDeltaEventType, F as resolveOpenAISupportedReasoningEfforts, H as stripUnsupportedSchemaKeywords, I as supportsOpenAIReasoningEffort, L as supportsOpenAITemperature, M as isOpenAIGpt56Model, N as normalizeOpenAIReasoningEffort, O as isResponsesTextDeltaEventType, P as resolveOpenAIReasoningEffortForModel, R as extractToolSchemaModelCompat, S as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, T as isAzureResponsesTextDeltaEvent, U as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, V as shouldOmitEmptyArrayItems, W as cleanSchemaForGemini, _ as readResponsesToolCallItemIdentity, b as resolveResponsesTerminalStopReason, c as clearOpenAIToolSchemaCacheForTest, d as normalizeOpenAIStrictToolParameters, f as normalizeStrictOpenAIJsonSchema, g as createResponsesToolCallTracker, h as normalizeOpenAIStrictCompatSchema, j as isOpenAIGpt55Model, k as resolveResponsesMessageSnapshotCollapse, l as findOpenAIStrictToolProjectionDiagnostics, m as findOpenAIStrictSchemaViolations, p as resolveOpenAIProjectedToolsStrictToolFlag, u as isStrictOpenAIJsonSchemaCompatible, v as mapResponsesTerminalUsage, w as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, x as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, y as readResponsesReasoningTokens, z as normalizeToolParameterSchema } from "../openai-responses-shared-CzCurmY1.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";
-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 };
+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-ITOU9bG1.mjs";
+import { n as streamOpenAIResponses, r as streamSimpleOpenAIResponses } from "../openai-responses-BHpmtUKo.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, findOpenAIStrictSchemaViolations, findOpenAIStrictToolProjectionDiagnostics, isAzureResponsesTextDeltaEvent, isAzureResponsesTextDeltaEventType, isOpenAICompatibleAzureResponsesBaseUrl, isOpenAIGpt54MiniModel, isOpenAIGpt55Model, isOpenAIGpt56Model, isResponsesTextContentPartType, isResponsesTextDeltaEventType, isStrictOpenAIJsonSchemaCompatible, isTraditionalAzureOpenAIHost, mapOpenAIStopReason, mapResponsesTerminalUsage, normalizeOpenAIReasoningEffort, normalizeOpenAIStrictCompatSchema, normalizeOpenAIStrictToolParameters, normalizeStrictOpenAIJsonSchema, normalizeToolParameterSchema, parseAzureDeploymentNameMap, projectOpenAITools, projectRuntimeToolInputSchema, readResponsesReasoningTokens, readResponsesToolCallItemIdentity, reconcileOpenAICompletionsToolChoice, reconcileOpenAIResponsesToolChoice, resolveAzureDeploymentNameFromMap, resolveOpenAIProjectedToolsStrictToolFlag, resolveOpenAIReasoningEffortForModel, resolveOpenAISupportedReasoningEfforts, resolveResponsesMessageSnapshotCollapse, resolveResponsesTerminalStopReason, 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-AFwwWium.mjs";
+import { a as RegisteredApiProvider, t as ApiProvider } from "../api-registry-BMphkFf1.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-BZiVbGwK.mjs";
 
 //#region packages/ai/src/internal/default-runtime.d.ts
@@ -23,7 +23,8 @@ declare const defaultLlmRuntime: {
   streamSimple: <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => AssistantMessageEventStreamContract;
   completeSimple: <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => Promise<AssistantMessage>;
 };
-declare const registerApiProvider: <TApi extends Api, TOptions extends StreamOptions>(provider: ApiProvider<TApi, TOptions>, sourceId?: string) => void, getApiProvider: (api: Api) => RegisteredApiProvider | undefined, getApiProviders: () => RegisteredApiProvider[], unregisterApiProviders: (sourceId: string) => void, clearApiProviders: () => void;
+declare const getApiProvider: (api: Api) => RegisteredApiProvider | undefined, getApiProviders: () => RegisteredApiProvider[];
+declare function clearApiProviders(): void;
 declare const stream: <TApi extends Api>(model: Model<TApi>, context: Context, options?: ProviderStreamOptions) => AssistantMessageEventStreamContract, complete: <TApi extends Api>(model: Model<TApi>, context: Context, options?: ProviderStreamOptions) => Promise<AssistantMessage>, streamSimple: <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => AssistantMessageEventStreamContract, completeSimple: <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => Promise<AssistantMessage>;
 //#endregion
 //#region packages/ai/src/env-api-keys.d.ts
@@ -139,10 +140,12 @@ declare function isConfiguredContextSizeOverflowError(errorMessage: string): boo
  * - 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. Pass contextWindow param to detect silent overflow.
+ *   sometimes returns rate limit errors instead of the explicit overflow error above. 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
@@ -242,4 +245,4 @@ type SseByteGuard = {
 };
 declare function createSseByteGuard(reader: ReadableStreamDefaultReader<Uint8Array>, opts: ReadSseStreamWithLimitOptions): SseByteGuard;
 //#endregion
-export { FirstStreamEventAbortController, FirstStreamEventInternalOptions, FirstStreamEventTimeoutContext, OpenAICodexJwtPayload, ReadSseStreamWithLimitOptions, ReasoningTagTextDelta, ReasoningTagTextPartitioner, SessionResourceCleanup, SseByteGuard, SseStreamOverflow, 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 };
+export { FirstStreamEventAbortController, FirstStreamEventInternalOptions, FirstStreamEventTimeoutContext, OpenAICodexJwtPayload, ReadSseStreamWithLimitOptions, ReasoningTagTextDelta, ReasoningTagTextPartitioner, SessionResourceCleanup, SseByteGuard, SseStreamOverflow, 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 };

package/dist/internal/runtime.mjs

@@ -1,14 +1,14 @@
-import { n as getEnvApiKey, t as findEnvKeys } from "../env-api-keys-DnhYpc27.mjs";
+import { n as getEnvApiKey, t as findEnvKeys } from "../env-api-keys-DrgeBuva.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 { t as sanitizeSurrogates } from "../sanitize-unicode-DT5o51ur.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";
-import { t as sanitizeSurrogates } from "../sanitize-unicode-DT5o51ur.mjs";
-import { t as createReasoningTagTextPartitioner } from "../reasoning-tag-text-partitioner-BlVe67g6.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 { i as registerSessionResourceCleanup, n as resolveOpenAICodexAccountId, r as cleanupSessionResources, t as decodeOpenAICodexJwtPayload } from "../openai-chatgpt-jwt-DhAAzLkj.mjs";
+import { t as shortHash } from "../hash-CHgqbJmD.mjs";
+import { t as createReasoningTagTextPartitioner } from "../reasoning-tag-text-partitioner-BlVe67g6.mjs";
+import { a as withFirstStreamEventTimeout, i as getFirstStreamEventTimeoutMs, n as createFirstStreamEventTimeoutError, r as getFirstStreamEventTimeoutHandler, t as createFirstStreamEventAbortController } from "../stream-first-event-timeout-DP4xEyBY.mjs";
 import { t as createSseByteGuard } from "../streaming-byte-guard-BrbkbwUu.mjs";
 //#region packages/ai/src/internal/default-runtime.ts
 const DEFAULT_RUNTIME_KEY = Symbol.for("openclaw.ai.defaultRuntime");
@@ -26,7 +26,10 @@ function resolveDefaultRuntime() {
 const defaultRuntime = resolveDefaultRuntime();
 const defaultApiRegistry = defaultRuntime.registry;
 const defaultLlmRuntime = defaultRuntime.runtime;
-const { registerApiProvider, getApiProvider, getApiProviders, unregisterApiProviders, clearApiProviders } = defaultApiRegistry;
+const { getApiProvider, getApiProviders } = defaultApiRegistry;
+function clearApiProviders() {
+	defaultApiRegistry.clearApiProviders();
+}
 const { stream, complete, streamSimple, completeSimple } = defaultLlmRuntime;
 //#endregion
 //#region packages/ai/src/utils/overflow.ts
@@ -59,7 +62,9 @@ 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: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow
+* - 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
 * - 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.
@@ -70,17 +75,20 @@ 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,
 	/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,
@@ -107,7 +115,7 @@ 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;
+	return message.usage.input + message.usage.cacheRead + message.usage.cacheWrite;
 }
 /**
 * Check if an assistant message represents a context overflow error.
@@ -133,10 +141,12 @@ 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. Pass contextWindow param to detect silent overflow.
+*   sometimes returns rate limit errors instead of the explicit overflow error above. 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
@@ -174,4 +184,4 @@ 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, registerApiProvider, registerSessionResourceCleanup, repairJson, resolveOpenAICodexAccountId, sanitizeSurrogates, shortHash, stream, streamSimple, unregisterApiProviders, 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, registerSessionResourceCleanup, repairJson, resolveOpenAICodexAccountId, sanitizeSurrogates, shortHash, stream, streamSimple, withFirstStreamEventTimeout };

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-AFwwWium.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 { C as isImageWithMediaPayload, D as buildBaseOptions, E as adjustMaxTokensForThinking, O as clampMaxTokensToModel, S as hasMediaPayload, a as stripSystemPromptCacheBoundary, b as extractToolResultBlockText, c as transformMessages, i as splitSystemPromptCacheBoundary, k as clampReasoning, n as ensureSystemPromptCacheBoundary, o as normalizePromptCapabilityIds, r as prependSystemPromptAdditionAfterCacheBoundary, s as normalizeStructuredPromptSection, t as SYSTEM_PROMPT_CACHE_BOUNDARY, x as extractToolResultText, y as describeToolResultMediaPlaceholder } from "../system-prompt-cache-boundary-CbHeV4_l.mjs";
 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 };
+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-D5Ps7Led.mjs}

@@ -1,10 +1,10 @@
-import { n as getEnvApiKey } from "./env-api-keys-DnhYpc27.mjs";
+import { n as getEnvApiKey } from "./env-api-keys-DrgeBuva.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 parseStreamingJson } from "./json-parse-BvXNt1-7.mjs";
+import { n as getAiTransportHost } from "./host-XYGZcgO8.mjs";
+import { C as isImageWithMediaPayload, D as buildBaseOptions, O as clampMaxTokensToModel, a as stripSystemPromptCacheBoundary, c as transformMessages, w as truncateUtf16Safe, x as extractToolResultText, y as describeToolResultMediaPlaceholder } from "./system-prompt-cache-boundary-CbHeV4_l.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 { 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 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;
 }