@openclaw/ai 2026.7.2-beta.6 → 2026.7.33

package/dist/internal/anthropic.d.mts

@@ -1,8 +1,58 @@
-import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, g as supportsClaudeNativeXhighEffort, h as supportsClaudeNativeMaxEffort, i as requiresClaudeDefaultSampling, l as resolveClaudeNativeThinkingLevelMap, o as resolveClaudeFable5ModelIdentity, p as supportsClaudeAdaptiveThinking, s as resolveClaudeModelIdentity, u as resolveClaudeOpus5ModelIdentity } from "../anthropic-SrGtwsJu.mjs";
+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 } from "../types-bzp5k29J.mjs";
-import { n as AnthropicOptions, r as AnthropicThinkingDisplay, t as AnthropicEffort } from "../provider-options-D8bB3z9b.mjs";
+import { E as Model, F as SimpleStreamOptions, R as StreamFunction, u as Context, z as StreamOptions } from "../types-DRgdPqaZ.mjs";
+import Anthropic from "@anthropic-ai/sdk";
+
 //#region packages/ai/src/providers/anthropic.d.ts
+type AnthropicEffort = "low" | "medium" | "high" | "xhigh" | "max";
+type AnthropicThinkingDisplay = "summarized" | "omitted";
+interface AnthropicOptions extends StreamOptions {
+  /**
+   * Enable extended thinking.
+   * For Opus 4.6+ and Sonnet 4.6: uses adaptive thinking (model decides when/how much to think).
+   * For older models: uses budget-based thinking with thinkingBudgetTokens.
+   */
+  thinkingEnabled?: boolean;
+  /**
+   * Token budget for extended thinking (older models only).
+   * Ignored for Opus 4.6+ and Sonnet 4.6, which use adaptive thinking.
+   */
+  thinkingBudgetTokens?: number;
+  /**
+   * Effort level for adaptive thinking (Opus 4.6+ and Sonnet 4.6).
+   * Controls how much thinking Claude allocates:
+   * - "max": Always thinks with no constraints (Opus 4.6 only)
+   * - "xhigh": Highest reasoning level (Opus 4.7+)
+   * - "high": Always thinks, deep reasoning (default)
+   * - "medium": Moderate thinking, may skip for simple queries
+   * - "low": Minimal thinking, skips for simple tasks
+   * Ignored for older models.
+   */
+  effort?: AnthropicEffort;
+  /**
+   * Controls how thinking content is returned in API responses.
+   * - "summarized": Thinking blocks contain summarized thinking text (default here).
+   * - "omitted": Thinking blocks return an empty thinking field; the encrypted
+   *   signature still travels back for multi-turn continuity. Use for faster
+   *   time-to-first-text-token when your UI does not surface thinking.
+   *
+   * Note: Anthropic's API default for Claude Opus 4.7+ and Claude Mythos Preview
+   * is "omitted". We default to "summarized" here to keep behavior consistent
+   * with older Claude 4 models. Set this explicitly to "omitted" to opt in.
+   */
+  thinkingDisplay?: AnthropicThinkingDisplay;
+  interleavedThinking?: boolean;
+  toolChoice?: "auto" | "any" | "none" | {
+    type: "tool";
+    name: string;
+  };
+  /**
+   * Pre-built Anthropic client instance. When provided, skips internal client
+   * construction entirely. Use this to inject alternative SDK clients such as
+   * `AnthropicVertex` that shares the same messaging API.
+   */
+  client?: Anthropic;
+}
 declare const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOptions>;
 type AnthropicSimpleStreamOptions = SimpleStreamOptions & {
   toolChoice?: AnthropicOptions["toolChoice"];
@@ -48,8 +98,8 @@ declare function defaultsClaudeAdaptiveThinking(model: {
   params?: Record<string, unknown>;
   api?: string;
 }): boolean;
-/** Remove unsupported assistant prefills while preserving completed tool-use turns. */
-declare function prepareClaudeNoPrefillRequestContext(model: Model, context: Context): Context;
+/** Remove Sonnet 5 assistant prefills while preserving completed tool-use turns. */
+declare function prepareClaudeSonnet5RequestContext(model: Model, context: Context): Context;
 declare function applyClaudeRequestContract(params: Record<string, unknown>, model: {
   id?: string;
   params?: Record<string, unknown>;
@@ -70,25 +120,21 @@ declare function applyAnthropicRefusal(output: AnthropicRefusalOutput, stopDetai
 //#endregion
 //#region packages/ai/src/providers/anthropic-server-fallback.d.ts
 /** Anthropic beta that re-serves safety refusals on an allowed fallback model. */
-declare const ANTHROPIC_SERVER_SIDE_FALLBACK_BETA = "server-side-fallback-2026-07-01";
-/** Let Anthropic select the recommended model for each refusal category. */
-declare const ANTHROPIC_SERVER_SIDE_FALLBACKS: "default";
-declare const CLAUDE_OPUS_FALLBACK_MODEL_COST: {
+declare const ANTHROPIC_SERVER_SIDE_FALLBACK_BETA = "server-side-fallback-2026-06-01";
+declare const CLAUDE_FABLE_5_FALLBACK_MODEL = "claude-opus-4-8";
+declare const CLAUDE_FABLE_5_FALLBACK_MODEL_COST: {
   readonly input: 5;
   readonly output: 25;
   readonly cacheRead: 0.5;
   readonly cacheWrite: 6.25;
 };
+declare function buildAnthropicServerSideFallbacks(): Array<{
+  model: string;
+}>;
 type AnthropicFallbackBoundary = {
   fromModel: string | null;
   toModel: string | null;
 };
-/** Resolve billed rates from the serving model reported by Anthropic's fallback stream. */
-declare function resolveAnthropicFallbackServingModelCost(params: {
-  requestedModelId: string;
-  servingModelId: string | null;
-  requestedCost: Model["cost"];
-}): Model["cost"];
 /** Reads a `fallback` content block marking where one model's output gives way to the next. */
 declare function readAnthropicFallbackBoundary(block: unknown): AnthropicFallbackBoundary | null;
 /**
@@ -162,13 +208,8 @@ 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;
@@ -187,8 +228,7 @@ 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_FALLBACKS, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, AnthropicCacheWriteUsage, type AnthropicEffort, AnthropicFallbackBoundary, AnthropicIterationUsageResult, AnthropicIterationUsageSnapshot, type AnthropicOptions, AnthropicProjectedToolChoice, AnthropicPromptUsageSnapshot, type AnthropicThinkingDisplay, AnthropicToolProjection, CLAUDE_OPUS_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeNoPrefillRequestContext, projectAnthropicTools, readAnthropicCacheWriteUsage, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveAnthropicFallbackServingModelCost, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeOpus5ModelIdentity, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };
+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 };

package/dist/internal/anthropic.mjs

@@ -1,5 +1,4 @@
-import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, g as supportsClaudeNativeXhighEffort, h as supportsClaudeNativeMaxEffort, i as requiresClaudeDefaultSampling, l as resolveClaudeNativeThinkingLevelMap, o as resolveClaudeFable5ModelIdentity, p as supportsClaudeAdaptiveThinking, s as resolveClaudeModelIdentity, u as resolveClaudeOpus5ModelIdentity } from "../src-QkygScBs.mjs";
-import { a as applyClaudeRequestContract, c as requiresClaudeAdaptiveThinking, d as usesClaudeStreamingRefusalContract, l as resolveModelBoundThinkingReplayMode, o as defaultsClaudeAdaptiveThinking, s as prepareClaudeNoPrefillRequestContext, u as usesClaudeFable5MessagesContract } from "../host-Dog2WQiR.mjs";
-import { _ as omitFoundryBearerCredentialHeaders, a as projectAnthropicTools, c as ANTHROPIC_OMITTED_REASONING_TEXT, d as ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, f as CLAUDE_OPUS_FALLBACK_MODEL_COST, g as applyAnthropicRefusal, h as resolveAnthropicFallbackServingModelCost, i as readLastAnthropicIterationUsage, l as findActiveAnthropicToolTurnAssistantIndex, m as readAnthropicFallbackBoundary, n as readAnthropicPromptUsageSnapshot, o as reconcileAnthropicToolChoice, p as applyAnthropicFallbackBoundary, r as readAnthropicUsageTokenCount, s as resolveOriginalAnthropicToolName, t as readAnthropicCacheWriteUsage, u as ANTHROPIC_SERVER_SIDE_FALLBACKS, v as usesFoundryBearerAuth } from "../anthropic-usage-DWU-x8MI.mjs";
-import { n as streamAnthropic, r as streamSimpleAnthropic } from "../anthropic-CH4UUnZr.mjs";
-export { ANTHROPIC_OMITTED_REASONING_TEXT, ANTHROPIC_SERVER_SIDE_FALLBACKS, ANTHROPIC_SERVER_SIDE_FALLBACK_BETA, CLAUDE_OPUS_FALLBACK_MODEL_COST, applyAnthropicFallbackBoundary, applyAnthropicRefusal, applyClaudeRequestContract, defaultsClaudeAdaptiveThinking, findActiveAnthropicToolTurnAssistantIndex, omitFoundryBearerCredentialHeaders, prepareClaudeNoPrefillRequestContext, projectAnthropicTools, readAnthropicCacheWriteUsage, readAnthropicFallbackBoundary, readAnthropicPromptUsageSnapshot, readAnthropicUsageTokenCount, readLastAnthropicIterationUsage, reconcileAnthropicToolChoice, requiresClaudeAdaptiveThinking, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveAnthropicFallbackServingModelCost, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeOpus5ModelIdentity, resolveClaudeSonnet5ModelIdentity, resolveModelBoundThinkingReplayMode, resolveOriginalAnthropicToolName, streamAnthropic, streamSimpleAnthropic, supportsClaudeAdaptiveThinking, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, usesClaudeFable5MessagesContract, usesClaudeStreamingRefusalContract, usesFoundryBearerAuth };
+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-CZ503MYJ.mjs";
+import { d as prepareClaudeSonnet5RequestContext, f as requiresClaudeAdaptiveThinking, h as usesClaudeStreamingRefusalContract, l as applyClaudeRequestContract, m as usesClaudeFable5MessagesContract, p as resolveModelBoundThinkingReplayMode, u as defaultsClaudeAdaptiveThinking } from "../transform-messages-BhGF_fF4.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-BML72tLz.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 };

package/dist/internal/openai.d.mts

@@ -1,11 +1,8 @@
-import { E as Model, F as SimpleStreamOptions, I as StopReason, R as StreamFunction, X as Usage, u as Context, z as StreamOptions } from "../types-bzp5k29J.mjs";
-import { _ as resolveOpenAIReasoningEffortForModel, a as OpenAICompletionsOptions, b as supportsOpenAITemperature, c as projectOpenAITools, d as OpenAIApiReasoningEffort, f as OpenAIReasoningEffort, g as normalizeOpenAIReasoningEffort, h as isOpenAIGpt56Model, l as reconcileOpenAICompletionsToolChoice, m as isOpenAIGpt55Model, o as OpenAICompletionsToolChoice, p as isOpenAIGpt54MiniModel, s as OpenAIToolProjection, u as reconcileOpenAIResponsesToolChoice, v as resolveOpenAISupportedReasoningEfforts, y as supportsOpenAIReasoningEffort } from "../provider-options-D8bB3z9b.mjs";
-import { t as ResolvedOpenAICompletionsCompat } from "../openai-completions-compat-DBWjXoMZ.mjs";
-import { n as clampOpenAIPromptCacheKey, t as OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH } from "../openai-prompt-cache-2uo_1OR1.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-DRgdPqaZ.mjs";
 import OpenAI from "openai";
 import { TSchema } from "typebox";
-import { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
 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
 /**
@@ -41,13 +38,7 @@ declare function normalizeToolParameterSchema(schema: unknown, options?: ToolPar
 //#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.
- */
+/** Resolves the Azure deployment name for a model id, falling back to the model id. */
 declare function resolveAzureDeploymentNameFromMap(params: {
   modelId: string;
   deploymentMap?: string;
@@ -61,24 +52,90 @@ declare function isOpenAICompatibleAzureResponsesBaseUrl(baseUrl: string): boole
 declare const GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS: Set<string>;
 declare function cleanSchemaForGemini(schema: unknown): TSchema;
 //#endregion
-//#region packages/ai/src/providers/clean-for-llamacpp-gbnf.d.ts
-/** llama.cpp rejects grammar repetitions whose expanded rule count reaches 2000. */
-declare const LLAMACPP_GBNF_MAX_REPETITION_THRESHOLD = 2000;
-/** Removes JSON Schema constraints that llama.cpp cannot compile into GBNF. */
-declare function cleanSchemaForLlamacppGbnf(schema: unknown): unknown;
-/** Reports schema paths that llama.cpp cannot compile into GBNF. */
-declare function findLlamacppGbnfSchemaViolations(schema: unknown, path: string): string[];
+//#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/openai-completions-messages.d.ts
-/** Convert a normalized transcript to OpenAI Chat Completions messages. */
+//#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-completions.d.ts
-declare const streamOpenAICompletions: StreamFunction<"openai-completions", OpenAICompletionsOptions>;
-declare const streamSimpleOpenAICompletions: StreamFunction<"openai-completions", SimpleStreamOptions>;
+//#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 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 {
@@ -130,71 +187,6 @@ declare function resolveResponsesMessageSnapshotCollapse(params: {
   nextPhase: string | undefined;
 }): ResponsesMessageSnapshotCollapse;
 //#endregion
-//#region packages/ai/src/providers/openai-responses-terminal-usage.d.ts
-/** Terminal usage payload, modeled structurally so untyped callers can pass raw records. */
-type ResponsesTerminalUsagePayload = {
-  input_tokens?: number | null;
-  output_tokens?: number | null;
-  total_tokens?: number | null;
-  input_tokens_details?: {
-    cached_tokens?: number | null;
-    cache_write_tokens?: number | null;
-  } | null;
-  output_tokens_details?: {
-    reasoning_tokens?: number | null;
-  } | null;
-};
-/**
- * Split a terminal usage payload into the priced buckets.
- *
- * OpenAI includes cache reads and writes in `input_tokens`, so both are subtracted out of the
- * billable input bucket. `total_tokens` comes from the payload, but never below the sum of the
- * split buckets: proxies routinely omit it (reporting 0 would understate the turn), and a payload
- * whose `cached_tokens` exceeds `input_tokens` clamps the input bucket, leaving the reported total
- * short of what the buckets actually price.
- */
-declare function mapResponsesTerminalUsage(usage: ResponsesTerminalUsagePayload | undefined | null): Pick<Usage, "input" | "output" | "cacheRead" | "cacheWrite" | "totalTokens"> | undefined;
-/** Reasoning tokens are reported by the agent path only; the package path does not track them. */
-declare function readResponsesReasoningTokens(usage: ResponsesTerminalUsagePayload | undefined | null): number | undefined;
-/**
- * Resolve the terminal stop reason, including the two overrides every Responses path shares: a
- * content-filtered turn is a provider error rather than a truncated answer, and a turn that
- * produced tool calls reports `toolUse` instead of a plain stop.
- */
-declare function resolveResponsesTerminalStopReason(params: {
-  status: OpenAI.Responses.ResponseStatus | undefined;
-  terminalEventType?: "response.completed" | "response.incomplete";
-  incompleteReason?: string;
-  hasToolCall: boolean;
-}): {
-  stopReason: StopReason;
-  errorMessage?: string;
-};
-//#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;
@@ -204,14 +196,6 @@ declare function mapOpenAIStopReason(reason: string | null, options?: {
   allowSingularToolCall?: boolean;
 }): OpenAIStopReasonResult;
 //#endregion
-//#region packages/ai/src/providers/openai-tool-schema-compat.d.ts
-/** Repairs recoverable OpenAI tool-schema shapes before canonical normalization. */
-declare function normalizeOpenAIStrictCompatSchema(schema: unknown): TSchema;
-/** Finds schema paths that violate OpenAI strict tool-schema requirements. */
-declare function findOpenAIStrictSchemaViolations(schema: unknown, path: string, options?: {
-  requireObjectRoot?: boolean;
-}): string[];
-//#endregion
 //#region packages/ai/src/providers/openai-tool-schema.d.ts
 /**
  * OpenAI strict-tool-schema normalization and diagnostics.
@@ -257,4 +241,4 @@ type RuntimeToolInputSchemaProjection = {
 /** 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, LLAMACPP_GBNF_MAX_REPETITION_THRESHOLD, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, OpenAIApiReasoningEffort, type OpenAICompletionsOptions, OpenAICompletionsToolChoice, OpenAIReasoningEffort, OpenAIResponsesOptions, OpenAIStopReasonResult, OpenAIToolProjection, ResponsesMessageSnapshotCollapse, ResponsesTerminalUsagePayload, ResponsesTextContentPartType, ResponsesTextDeltaEventType, ResponsesToolCallIdentity, ResponsesToolCallState, RuntimeToolInputSchemaJson, RuntimeToolInputSchemaProjection, ToolParameterSchemaOptions, ToolSchemaModelCompat, clampOpenAIPromptCacheKey, cleanSchemaForGemini, cleanSchemaForLlamacppGbnf, clearOpenAIToolSchemaCacheForTest, convertMessages, createResponsesToolCallTracker, extractToolSchemaModelCompat, findLlamacppGbnfSchemaViolations, 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 };
+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, RuntimeToolInputSchemaJson, RuntimeToolInputSchemaProjection, ToolParameterSchemaOptions, ToolSchemaModelCompat, clampOpenAIPromptCacheKey, cleanSchemaForGemini, clearOpenAIToolSchemaCacheForTest, convertMessages, extractToolSchemaModelCompat, findOpenAIStrictToolProjectionDiagnostics, isAzureResponsesTextDeltaEvent, isAzureResponsesTextDeltaEventType, isOpenAICompatibleAzureResponsesBaseUrl, isOpenAIGpt54MiniModel, isOpenAIGpt55Model, isOpenAIGpt56Model, isResponsesTextContentPartType, isResponsesTextDeltaEventType, isStrictOpenAIJsonSchemaCompatible, isTraditionalAzureOpenAIHost, mapOpenAIStopReason, normalizeOpenAIReasoningEffort, normalizeOpenAIStrictToolParameters, normalizeStrictOpenAIJsonSchema, normalizeToolParameterSchema, parseAzureDeploymentNameMap, projectOpenAITools, projectRuntimeToolInputSchema, reconcileOpenAICompletionsToolChoice, reconcileOpenAIResponsesToolChoice, resolveAzureDeploymentNameFromMap, resolveOpenAIProjectedToolsStrictToolFlag, resolveOpenAIReasoningEffortForModel, resolveOpenAISupportedReasoningEfforts, resolveResponsesMessageSnapshotCollapse, resolveUnsupportedToolSchemaKeywords, shouldOmitEmptyArrayItems, streamOpenAICompletions, streamOpenAIResponses, streamSimpleOpenAICompletions, streamSimpleOpenAIResponses, stripUnsupportedSchemaKeywords, supportsOpenAIReasoningEffort };

package/dist/internal/openai.mjs

@@ -1,9 +1,7 @@
-import { n as clampOpenAIPromptCacheKey, t as OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH } from "../openai-prompt-cache-mZTCdRPo.mjs";
-import { n as reconcileOpenAICompletionsToolChoice, r as reconcileOpenAIResponsesToolChoice, t as projectOpenAITools } from "../openai-tool-projection-OhX64DoP.mjs";
 import { t as projectRuntimeToolInputSchema } from "../tool-schema-json-projection-BwNu3nDi.mjs";
-import { n as mapOpenAIStopReason, r as convertMessages } from "../openai-reasoning-compat-YgeLncHw.mjs";
-import { $ as normalizeOpenAIStrictCompatSchema, J as isStrictOpenAIJsonSchemaCompatible, K as clearOpenAIToolSchemaCacheForTest, Q as findOpenAIStrictSchemaViolations, X as normalizeStrictOpenAIJsonSchema, Y as normalizeOpenAIStrictToolParameters, Z as resolveOpenAIProjectedToolsStrictToolFlag, _ as resolveResponsesMessageSnapshotCollapse, _t as supportsOpenAITemperature, a as resolveResponsesTerminalStopReason, at as LLAMACPP_GBNF_MAX_REPETITION_THRESHOLD, c as readResponsesToolCallItemIdentity, ct as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, d as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, dt as isOpenAIGpt55Model, et as extractToolSchemaModelCompat, f as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, ft as isOpenAIGpt56Model, g as isResponsesTextDeltaEventType, gt as supportsOpenAIReasoningEffort, h as isResponsesTextContentPartType, ht as resolveOpenAISupportedReasoningEfforts, i as readResponsesReasoningTokens, it as stripUnsupportedSchemaKeywords, l as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, lt as cleanSchemaForGemini, m as isAzureResponsesTextDeltaEventType, mt as resolveOpenAIReasoningEffortForModel, nt as resolveUnsupportedToolSchemaKeywords, ot as cleanSchemaForLlamacppGbnf, p as isAzureResponsesTextDeltaEvent, pt as normalizeOpenAIReasoningEffort, q as findOpenAIStrictToolProjectionDiagnostics, r as mapResponsesTerminalUsage, rt as shouldOmitEmptyArrayItems, s as createResponsesToolCallTracker, st as findLlamacppGbnfSchemaViolations, tt as normalizeToolParameterSchema, u as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, ut as isOpenAIGpt54MiniModel } from "../openai-responses-stream-internal-Cw5txaGW.mjs";
-import { i as resolveAzureDeploymentNameFromMap, n as isTraditionalAzureOpenAIHost, r as parseAzureDeploymentNameMap, t as isOpenAICompatibleAzureResponsesBaseUrl } from "../azure-openai-responses-client-compat-C7K7QfUE.mjs";
-import { n as streamOpenAICompletions, r as streamSimpleOpenAICompletions } from "../openai-completions-DsOxhOD1.mjs";
-import { n as streamOpenAIResponses, r as streamSimpleOpenAIResponses } from "../openai-responses-BT7A3sLu.mjs";
-export { AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, LLAMACPP_GBNF_MAX_REPETITION_THRESHOLD, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH, OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, clampOpenAIPromptCacheKey, cleanSchemaForGemini, cleanSchemaForLlamacppGbnf, clearOpenAIToolSchemaCacheForTest, convertMessages, createResponsesToolCallTracker, extractToolSchemaModelCompat, findLlamacppGbnfSchemaViolations, 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 };
+import { A as extractToolSchemaModelCompat, C as isOpenAIGpt54MiniModel, D as resolveOpenAIReasoningEffortForModel, E as normalizeOpenAIReasoningEffort, F as GEMINI_UNSUPPORTED_SCHEMA_KEYWORDS, I as cleanSchemaForGemini, M as resolveUnsupportedToolSchemaKeywords, N as shouldOmitEmptyArrayItems, O as resolveOpenAISupportedReasoningEfforts, P as stripUnsupportedSchemaKeywords, S as resolveResponsesMessageSnapshotCollapse, T as isOpenAIGpt56Model, _ as OPENAI_RESPONSES_OUTPUT_TEXT_DELTA_EVENT_TYPE, b as isResponsesTextContentPartType, c as clearOpenAIToolSchemaCacheForTest, d as normalizeOpenAIStrictToolParameters, f as normalizeStrictOpenAIJsonSchema, g as OPENAI_RESPONSES_OUTPUT_TEXT_CONTENT_PART_TYPE, h as AZURE_RESPONSES_TEXT_DELTA_EVENT_TYPE, j as normalizeToolParameterSchema, k as supportsOpenAIReasoningEffort, l as findOpenAIStrictToolProjectionDiagnostics, m as AZURE_RESPONSES_TEXT_CONTENT_PART_TYPE, p as resolveOpenAIProjectedToolsStrictToolFlag, u as isStrictOpenAIJsonSchemaCompatible, v as isAzureResponsesTextDeltaEvent, w as isOpenAIGpt55Model, x as isResponsesTextDeltaEventType, y as isAzureResponsesTextDeltaEventType } from "../openai-responses-shared-L3PWZvkm.mjs";
+import { i as resolveAzureDeploymentNameFromMap, n as isTraditionalAzureOpenAIHost, r as parseAzureDeploymentNameMap, t as isOpenAICompatibleAzureResponsesBaseUrl } from "../azure-openai-responses-client-compat-a_O_GVQV.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-BPlyXt8H.mjs";
+import { a as mapOpenAIStopReason, i as streamSimpleOpenAICompletions, r as streamOpenAICompletions, t as convertMessages } from "../openai-completions-9qYFVS7H.mjs";
+import { n as streamOpenAIResponses, r as streamSimpleOpenAIResponses } from "../openai-responses-D6l9URyV.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, extractToolSchemaModelCompat, findOpenAIStrictToolProjectionDiagnostics, isAzureResponsesTextDeltaEvent, isAzureResponsesTextDeltaEventType, isOpenAICompatibleAzureResponsesBaseUrl, isOpenAIGpt54MiniModel, isOpenAIGpt55Model, isOpenAIGpt56Model, isResponsesTextContentPartType, isResponsesTextDeltaEventType, isStrictOpenAIJsonSchemaCompatible, isTraditionalAzureOpenAIHost, mapOpenAIStopReason, normalizeOpenAIReasoningEffort, normalizeOpenAIStrictToolParameters, normalizeStrictOpenAIJsonSchema, normalizeToolParameterSchema, parseAzureDeploymentNameMap, projectOpenAITools, projectRuntimeToolInputSchema, reconcileOpenAICompletionsToolChoice, reconcileOpenAIResponsesToolChoice, resolveAzureDeploymentNameFromMap, resolveOpenAIProjectedToolsStrictToolFlag, resolveOpenAIReasoningEffortForModel, resolveOpenAISupportedReasoningEfforts, resolveResponsesMessageSnapshotCollapse, resolveUnsupportedToolSchemaKeywords, shouldOmitEmptyArrayItems, streamOpenAICompletions, streamOpenAIResponses, streamSimpleOpenAICompletions, streamSimpleOpenAIResponses, stripUnsupportedSchemaKeywords, supportsOpenAIReasoningEffort };

package/dist/internal/runtime.d.mts

@@ -1,7 +1,6 @@
-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-bzp5k29J.mjs";
-import { a as RegisteredApiProvider, t as ApiProvider } from "../api-registry-DlMgPR39.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-DRgdPqaZ.mjs";
+import { a as RegisteredApiProvider, t as ApiProvider } from "../api-registry-BXYnCOIR.mjs";
 import { t as sanitizeSurrogates } from "../sanitize-unicode-BZiVbGwK.mjs";
-import { a as createFirstStreamEventTimeoutError, c as withFirstStreamEventTimeout, i as createFirstStreamEventAbortController, n as FirstStreamEventInternalOptions, o as getFirstStreamEventTimeoutHandler, r as FirstStreamEventTimeoutContext, s as getFirstStreamEventTimeoutMs, t as FirstStreamEventAbortController } from "../stream-first-event-timeout-DvDeSucC.mjs";
 
 //#region packages/ai/src/internal/default-runtime.d.ts
 declare const defaultApiRegistry: {
@@ -24,8 +23,7 @@ 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 getApiProvider: (api: Api) => RegisteredApiProvider | undefined, getApiProviders: () => RegisteredApiProvider[];
-declare function clearApiProviders(): void;
+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 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
@@ -141,12 +139,10 @@ 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 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
@@ -170,7 +166,7 @@ declare function isConfiguredContextSizeOverflowError(errorMessage: string): boo
  */
 declare function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean;
 //#endregion
-//#region packages/markdown-core/src/reasoning-tag-parser.d.ts
+//#region packages/ai/src/utils/reasoning-tag-text-partitioner.d.ts
 type ReasoningTagTextDelta = {
   kind: "text";
   text: string;
@@ -178,8 +174,6 @@ type ReasoningTagTextDelta = {
   kind: "thinking";
   text: string;
 };
-//#endregion
-//#region packages/markdown-core/src/reasoning-tags.d.ts
 interface ReasoningTagTextPartitioner {
   markStrict(): void;
   push(chunk: string): ReasoningTagTextDelta[];
@@ -188,9 +182,36 @@ interface ReasoningTagTextPartitioner {
   hasPending(): boolean;
   isInsideReasoning(): boolean;
 }
-/** Creates a block-incremental parser that emits only Markdown-stable text. */
 declare function createReasoningTagTextPartitioner(): ReasoningTagTextPartitioner;
 //#endregion
+//#region packages/ai/src/utils/stream-first-event-timeout.d.ts
+type StreamStage = "responses" | "completions";
+type FirstStreamEventTimeoutContext = {
+  provider?: string;
+  api?: string;
+  model?: string;
+  timeoutMs: number;
+  stage?: StreamStage;
+  hint?: string;
+  abort?: (reason: Error) => void;
+  onTimeout?: (reason: Error) => void;
+};
+type FirstStreamEventInternalOptions = {
+  firstEventTimeoutMs?: number;
+  abortFirstEventStream?: (reason: Error) => void;
+  onFirstEventTimeout?: (reason: Error) => void;
+};
+type FirstStreamEventAbortController = {
+  signal: AbortSignal;
+  abort: (reason: Error) => void;
+  dispose: () => void;
+};
+declare function getFirstStreamEventTimeoutMs(options: unknown): number | undefined;
+declare function getFirstStreamEventTimeoutHandler(options: unknown): ((reason: Error) => void) | undefined;
+declare function createFirstStreamEventTimeoutError(context: FirstStreamEventTimeoutContext): Error;
+declare function createFirstStreamEventAbortController(parentSignal?: AbortSignal): FirstStreamEventAbortController;
+declare function withFirstStreamEventTimeout<T>(stream: AsyncIterable<T>, context: FirstStreamEventTimeoutContext): AsyncIterable<T>;
+//#endregion
 //#region packages/ai/src/utils/streaming-byte-guard.d.ts
 /**
  * Bounded SSE / NDJSON stream reader guard.
@@ -221,4 +242,4 @@ type SseByteGuard = {
 };
 declare function createSseByteGuard(reader: ReadableStreamDefaultReader<Uint8Array>, opts: ReadSseStreamWithLimitOptions): SseByteGuard;
 //#endregion
-export { FirstStreamEventAbortController, FirstStreamEventInternalOptions, FirstStreamEventTimeoutContext, OpenAICodexJwtPayload, ReadSseStreamWithLimitOptions, type ReasoningTagTextDelta, type 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 };
+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 };

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-DgmOla96.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 };