@dexto/core 1.5.0 → 1.5.1

package/dist/llm/executor/provider-options.cjs

@@ -0,0 +1,87 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var provider_options_exports = {};
+__export(provider_options_exports, {
+  buildProviderOptions: () => buildProviderOptions,
+  getDefaultReasoningEffort: () => getDefaultReasoningEffort
+});
+module.exports = __toCommonJS(provider_options_exports);
+var import_registry = require("../registry.js");
+function buildProviderOptions(config) {
+  const { provider, model, reasoningEffort } = config;
+  const modelLower = model.toLowerCase();
+  if (provider === "anthropic") {
+    return {
+      anthropic: {
+        // Enable prompt caching - saves money and improves latency
+        cacheControl: { type: "ephemeral" },
+        // Stream reasoning/thinking content when model supports it
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "bedrock" && modelLower.includes("claude")) {
+    return {
+      bedrock: {
+        cacheControl: { type: "ephemeral" },
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "vertex" && modelLower.includes("claude")) {
+    return {
+      "vertex-anthropic": {
+        cacheControl: { type: "ephemeral" },
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "google" || provider === "vertex" && !modelLower.includes("claude")) {
+    return {
+      google: {
+        thinkingConfig: {
+          // Include thoughts in the response for transparency
+          includeThoughts: true
+        }
+      }
+    };
+  }
+  if (provider === "openai") {
+    const effectiveEffort = reasoningEffort ?? getDefaultReasoningEffort(model);
+    if (effectiveEffort) {
+      return {
+        openai: {
+          reasoningEffort: effectiveEffort
+        }
+      };
+    }
+  }
+  return void 0;
+}
+function getDefaultReasoningEffort(model) {
+  if ((0, import_registry.isReasoningCapableModel)(model)) {
+    return "medium";
+  }
+  return void 0;
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  buildProviderOptions,
+  getDefaultReasoningEffort
+});

package/dist/llm/executor/provider-options.d.ts

@@ -0,0 +1,49 @@
+/**
+ * Provider-specific options builder for Vercel AI SDK's streamText/generateText.
+ *
+ * Centralizes provider-specific configuration that requires explicit opt-in:
+ * - Anthropic: cacheControl for prompt caching, sendReasoning for extended thinking
+ * - Bedrock/Vertex Claude: Same as Anthropic (Claude models on these platforms)
+ * - Google: thinkingConfig for Gemini thinking models
+ * - OpenAI: reasoningEffort for o1/o3/codex/gpt-5 models
+ *
+ * Caching notes:
+ * - Anthropic: Requires explicit cacheControl option (we enable it)
+ * - OpenAI: Automatic for prompts ≥1024 tokens (no config needed)
+ * - Google: Implicit caching automatic for Gemini 2.5+ (≥1024 tokens for Flash,
+ *   ≥2048 for Pro). Explicit caching requires pre-created cachedContent IDs.
+ * All providers return cached token counts in the response (cachedInputTokens).
+ */
+import type { LLMProvider } from '../types.js';
+export type ReasoningEffort = 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh';
+export interface ProviderOptionsConfig {
+    provider: LLMProvider;
+    model: string;
+    reasoningEffort?: ReasoningEffort | undefined;
+}
+/**
+ * Build provider-specific options for streamText/generateText.
+ *
+ * @param config Provider, model, and optional reasoning effort configuration
+ * @returns Provider options object or undefined if no special options needed
+ */
+export declare function buildProviderOptions(config: ProviderOptionsConfig): Record<string, Record<string, unknown>> | undefined;
+/**
+ * Determine the default reasoning effort for OpenAI models.
+ *
+ * OpenAI reasoning effort levels (from lowest to highest):
+ * - 'none': No reasoning, fastest responses
+ * - 'low': Minimal reasoning, fast responses
+ * - 'medium': Balanced reasoning (OpenAI's recommended daily driver)
+ * - 'high': Thorough reasoning for complex tasks
+ * - 'xhigh': Extra high reasoning for quality-critical, non-latency-sensitive tasks
+ *
+ * Default strategy:
+ * - Reasoning-capable models (codex, o1, o3, gpt-5): 'medium' - OpenAI's recommended default
+ * - Other models: undefined (no reasoning effort needed)
+ *
+ * @param model The model name
+ * @returns Reasoning effort level or undefined if not applicable
+ */
+export declare function getDefaultReasoningEffort(model: string): Exclude<ReasoningEffort, 'none'> | undefined;
+//# sourceMappingURL=provider-options.d.ts.map

package/dist/llm/executor/provider-options.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider-options.d.ts","sourceRoot":"","sources":["../../../src/llm/executor/provider-options.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAG/C,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAEvF,MAAM,WAAW,qBAAqB;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,eAAe,CAAC,EAAE,eAAe,GAAG,SAAS,CAAC;CACjD;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAChC,MAAM,EAAE,qBAAqB,GAC9B,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,GAAG,SAAS,CAgErD;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,yBAAyB,CACrC,KAAK,EAAE,MAAM,GACd,OAAO,CAAC,eAAe,EAAE,MAAM,CAAC,GAAG,SAAS,CAS9C"}

package/dist/llm/executor/provider-options.js

@@ -0,0 +1,63 @@
+import "../../chunk-PTJYTZNU.js";
+import { isReasoningCapableModel } from "../registry.js";
+function buildProviderOptions(config) {
+  const { provider, model, reasoningEffort } = config;
+  const modelLower = model.toLowerCase();
+  if (provider === "anthropic") {
+    return {
+      anthropic: {
+        // Enable prompt caching - saves money and improves latency
+        cacheControl: { type: "ephemeral" },
+        // Stream reasoning/thinking content when model supports it
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "bedrock" && modelLower.includes("claude")) {
+    return {
+      bedrock: {
+        cacheControl: { type: "ephemeral" },
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "vertex" && modelLower.includes("claude")) {
+    return {
+      "vertex-anthropic": {
+        cacheControl: { type: "ephemeral" },
+        sendReasoning: true
+      }
+    };
+  }
+  if (provider === "google" || provider === "vertex" && !modelLower.includes("claude")) {
+    return {
+      google: {
+        thinkingConfig: {
+          // Include thoughts in the response for transparency
+          includeThoughts: true
+        }
+      }
+    };
+  }
+  if (provider === "openai") {
+    const effectiveEffort = reasoningEffort ?? getDefaultReasoningEffort(model);
+    if (effectiveEffort) {
+      return {
+        openai: {
+          reasoningEffort: effectiveEffort
+        }
+      };
+    }
+  }
+  return void 0;
+}
+function getDefaultReasoningEffort(model) {
+  if (isReasoningCapableModel(model)) {
+    return "medium";
+  }
+  return void 0;
+}
+export {
+  buildProviderOptions,
+  getDefaultReasoningEffort
+};

package/dist/llm/executor/stream-processor.cjs

@@ -158,8 +158,10 @@ class StreamProcessor {
           }
           case "finish-step":
             if (event.usage) {
-              const cacheWriteTokens = event.providerMetadata?.["anthropic"]?.["cacheCreationInputTokens"] ?? // @ts-expect-error - Bedrock metadata typing not in Vercel SDK
-              event.providerMetadata?.["bedrock"]?.["usage"]?.["cacheWriteInputTokens"] ?? 0;
+              const anthropicMeta = event.providerMetadata?.["anthropic"];
+              const bedrockMeta = event.providerMetadata?.["bedrock"];
+              const cacheWriteTokens = anthropicMeta?.["cacheCreationInputTokens"] ?? bedrockMeta?.usage?.["cacheWriteInputTokens"] ?? 0;
+              const cacheReadTokens = anthropicMeta?.["cacheReadInputTokens"] ?? bedrockMeta?.usage?.["cacheReadInputTokens"] ?? event.usage.cachedInputTokens ?? 0;
               this.actualTokens = {
                 inputTokens: (this.actualTokens.inputTokens ?? 0) + (event.usage.inputTokens ?? 0),
                 outputTokens: (this.actualTokens.outputTokens ?? 0) + (event.usage.outputTokens ?? 0),
@@ -168,16 +170,17 @@ class StreamProcessor {
                   reasoningTokens: (this.actualTokens.reasoningTokens ?? 0) + event.usage.reasoningTokens
                 },
                 // Cache tokens
-                cacheReadTokens: (this.actualTokens.cacheReadTokens ?? 0) + (event.usage.cachedInputTokens ?? 0),
+                cacheReadTokens: (this.actualTokens.cacheReadTokens ?? 0) + cacheReadTokens,
                 cacheWriteTokens: (this.actualTokens.cacheWriteTokens ?? 0) + cacheWriteTokens
               };
             }
             break;
           case "finish": {
             this.finishReason = event.finishReason;
-            const cachedInputTokens = event.totalUsage.cachedInputTokens ?? 0;
+            const cacheReadTokens = this.actualTokens.cacheReadTokens ?? event.totalUsage.cachedInputTokens ?? 0;
+            const cacheWriteTokens = this.actualTokens.cacheWriteTokens ?? 0;
             const providerExcludesCached = this.config.provider === "anthropic" || this.config.provider === "bedrock";
-            const adjustedInputTokens = providerExcludesCached ? event.totalUsage.inputTokens ?? 0 : (event.totalUsage.inputTokens ?? 0) - cachedInputTokens;
+            const adjustedInputTokens = providerExcludesCached ? event.totalUsage.inputTokens ?? 0 : (event.totalUsage.inputTokens ?? 0) - cacheReadTokens;
             const usage = {
               inputTokens: adjustedInputTokens,
               outputTokens: event.totalUsage.outputTokens ?? 0,
@@ -186,9 +189,9 @@ class StreamProcessor {
               ...event.totalUsage.reasoningTokens !== void 0 && {
                 reasoningTokens: event.totalUsage.reasoningTokens
               },
-              // Cache tokens - read from totalUsage, write from accumulated finish-step events
-              cacheReadTokens: cachedInputTokens,
-              cacheWriteTokens: this.actualTokens.cacheWriteTokens ?? 0
+              // Cache tokens from accumulated finish-step events or totalUsage fallback
+              cacheReadTokens,
+              cacheWriteTokens
             };
             this.actualTokens = usage;
             this.logger.info("LLM response complete", {

package/dist/llm/executor/stream-processor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stream-processor.d.ts","sourceRoot":"","sources":["../../../src/llm/executor/stream-processor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,OAAO,IAAI,aAAa,EAAE,MAAM,IAAI,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAC1D,OAAO,EAAE,eAAe,EAAmB,MAAM,uBAAuB,CAAC;AACzE,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAE3D,OAAO,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AAGnD,OAAO,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAExD,OAAO,EAAE,WAAW,EAAc,MAAM,aAAa,CAAC;AAEtD,MAAM,WAAW,qBAAqB;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,eAAe;IAwBpB,OAAO,CAAC,cAAc;IACtB,OAAO,CAAC,QAAQ;IAChB,OAAO,CAAC,eAAe;IACvB,OAAO,CAAC,WAAW;IACnB,OAAO,CAAC,MAAM;IAEd,OAAO,CAAC,SAAS;IACjB,OAAO,CAAC,gBAAgB,CAAC;IA9B7B,OAAO,CAAC,kBAAkB,CAAuB;IACjD,OAAO,CAAC,YAAY,CAAmE;IACvF,OAAO,CAAC,YAAY,CAA8B;IAClD,OAAO,CAAC,aAAa,CAAc;IACnC,OAAO,CAAC,eAAe,CAAc;IACrC,OAAO,CAAC,MAAM,CAAe;IAC7B;;;OAGG;IACH,OAAO,CAAC,gBAAgB,CAAgD;IAExE;;;;;;;;;OASG;gBAES,cAAc,EAAE,cAAc,EAC9B,QAAQ,EAAE,eAAe,EACzB,eAAe,EAAE,eAAe,EAChC,WAAW,EAAE,WAAW,EACxB,MAAM,EAAE,qBAAqB,EACrC,MAAM,EAAE,YAAY,EACZ,SAAS,GAAE,OAAc,EACzB,gBAAgB,CAAC,EAAE,GAAG,CAC1B,MAAM,EACN;QAAE,eAAe,EAAE,OAAO,CAAC;QAAC,cAAc,CAAC,EAAE,UAAU,GAAG,UAAU,CAAA;KAAE,CACzE,YAAA;IAKC,OAAO,CACT,QAAQ,EAAE,MAAM,gBAAgB,CAAC,aAAa,EAAE,OAAO,CAAC,GACzD,OAAO,CAAC,qBAAqB,CAAC;YAoZnB,sBAAsB;YAKtB,gBAAgB;IAO9B;;;;OAIG;YACW,2BAA2B;CAmC5C"}
+{"version":3,"file":"stream-processor.d.ts","sourceRoot":"","sources":["../../../src/llm/executor/stream-processor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,OAAO,IAAI,aAAa,EAAE,MAAM,IAAI,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAC1D,OAAO,EAAE,eAAe,EAAmB,MAAM,uBAAuB,CAAC;AACzE,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAE3D,OAAO,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AAGnD,OAAO,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAExD,OAAO,EAAE,WAAW,EAAc,MAAM,aAAa,CAAC;AAEtD,MAAM,WAAW,qBAAqB;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,eAAe;IAwBpB,OAAO,CAAC,cAAc;IACtB,OAAO,CAAC,QAAQ;IAChB,OAAO,CAAC,eAAe;IACvB,OAAO,CAAC,WAAW;IACnB,OAAO,CAAC,MAAM;IAEd,OAAO,CAAC,SAAS;IACjB,OAAO,CAAC,gBAAgB,CAAC;IA9B7B,OAAO,CAAC,kBAAkB,CAAuB;IACjD,OAAO,CAAC,YAAY,CAAmE;IACvF,OAAO,CAAC,YAAY,CAA8B;IAClD,OAAO,CAAC,aAAa,CAAc;IACnC,OAAO,CAAC,eAAe,CAAc;IACrC,OAAO,CAAC,MAAM,CAAe;IAC7B;;;OAGG;IACH,OAAO,CAAC,gBAAgB,CAAgD;IAExE;;;;;;;;;OASG;gBAES,cAAc,EAAE,cAAc,EAC9B,QAAQ,EAAE,eAAe,EACzB,eAAe,EAAE,eAAe,EAChC,WAAW,EAAE,WAAW,EACxB,MAAM,EAAE,qBAAqB,EACrC,MAAM,EAAE,YAAY,EACZ,SAAS,GAAE,OAAc,EACzB,gBAAgB,CAAC,EAAE,GAAG,CAC1B,MAAM,EACN;QAAE,eAAe,EAAE,OAAO,CAAC;QAAC,cAAc,CAAC,EAAE,UAAU,GAAG,UAAU,CAAA;KAAE,CACzE,YAAA;IAKC,OAAO,CACT,QAAQ,EAAE,MAAM,gBAAgB,CAAC,aAAa,EAAE,OAAO,CAAC,GACzD,OAAO,CAAC,qBAAqB,CAAC;YAmanB,sBAAsB;YAKtB,gBAAgB;IAO9B;;;;OAIG;YACW,2BAA2B;CAmC5C"}

package/dist/llm/executor/stream-processor.js

@@ -136,8 +136,10 @@ class StreamProcessor {
           }
           case "finish-step":
             if (event.usage) {
-              const cacheWriteTokens = event.providerMetadata?.["anthropic"]?.["cacheCreationInputTokens"] ?? // @ts-expect-error - Bedrock metadata typing not in Vercel SDK
-              event.providerMetadata?.["bedrock"]?.["usage"]?.["cacheWriteInputTokens"] ?? 0;
+              const anthropicMeta = event.providerMetadata?.["anthropic"];
+              const bedrockMeta = event.providerMetadata?.["bedrock"];
+              const cacheWriteTokens = anthropicMeta?.["cacheCreationInputTokens"] ?? bedrockMeta?.usage?.["cacheWriteInputTokens"] ?? 0;
+              const cacheReadTokens = anthropicMeta?.["cacheReadInputTokens"] ?? bedrockMeta?.usage?.["cacheReadInputTokens"] ?? event.usage.cachedInputTokens ?? 0;
               this.actualTokens = {
                 inputTokens: (this.actualTokens.inputTokens ?? 0) + (event.usage.inputTokens ?? 0),
                 outputTokens: (this.actualTokens.outputTokens ?? 0) + (event.usage.outputTokens ?? 0),
@@ -146,16 +148,17 @@ class StreamProcessor {
                   reasoningTokens: (this.actualTokens.reasoningTokens ?? 0) + event.usage.reasoningTokens
                 },
                 // Cache tokens
-                cacheReadTokens: (this.actualTokens.cacheReadTokens ?? 0) + (event.usage.cachedInputTokens ?? 0),
+                cacheReadTokens: (this.actualTokens.cacheReadTokens ?? 0) + cacheReadTokens,
                 cacheWriteTokens: (this.actualTokens.cacheWriteTokens ?? 0) + cacheWriteTokens
               };
             }
             break;
           case "finish": {
             this.finishReason = event.finishReason;
-            const cachedInputTokens = event.totalUsage.cachedInputTokens ?? 0;
+            const cacheReadTokens = this.actualTokens.cacheReadTokens ?? event.totalUsage.cachedInputTokens ?? 0;
+            const cacheWriteTokens = this.actualTokens.cacheWriteTokens ?? 0;
             const providerExcludesCached = this.config.provider === "anthropic" || this.config.provider === "bedrock";
-            const adjustedInputTokens = providerExcludesCached ? event.totalUsage.inputTokens ?? 0 : (event.totalUsage.inputTokens ?? 0) - cachedInputTokens;
+            const adjustedInputTokens = providerExcludesCached ? event.totalUsage.inputTokens ?? 0 : (event.totalUsage.inputTokens ?? 0) - cacheReadTokens;
             const usage = {
               inputTokens: adjustedInputTokens,
               outputTokens: event.totalUsage.outputTokens ?? 0,
@@ -164,9 +167,9 @@ class StreamProcessor {
               ...event.totalUsage.reasoningTokens !== void 0 && {
                 reasoningTokens: event.totalUsage.reasoningTokens
               },
-              // Cache tokens - read from totalUsage, write from accumulated finish-step events
-              cacheReadTokens: cachedInputTokens,
-              cacheWriteTokens: this.actualTokens.cacheWriteTokens ?? 0
+              // Cache tokens from accumulated finish-step events or totalUsage fallback
+              cacheReadTokens,
+              cacheWriteTokens
             };
             this.actualTokens = usage;
             this.logger.info("LLM response complete", {

package/dist/llm/executor/turn-executor.cjs

@@ -79,6 +79,7 @@ module.exports = __toCommonJS(turn_executor_exports);
 var import_ai = require("ai");
 var import_api = require("@opentelemetry/api");
 var import_stream_processor = require("./stream-processor.js");
+var import_provider_options = require("./provider-options.js");
 var import_types4 = require("../../logger/v2/types.js");
 var import_defer = require("../../utils/defer.js");
 var import_DextoRuntimeError = require("../../errors/DextoRuntimeError.js");
@@ -203,6 +204,11 @@ class TurnExecutor {
             streaming,
             this.approvalMetadata
           );
+          const providerOptions = (0, import_provider_options.buildProviderOptions)({
+            provider: this.llmContext.provider,
+            model: this.llmContext.model,
+            reasoningEffort: this.config.reasoningEffort
+          });
           const result = await streamProcessor.process(
             () => (0, import_ai.streamText)({
               model: this.model,
@@ -216,6 +222,10 @@ class TurnExecutor {
               ...this.config.temperature !== void 0 && {
                 temperature: this.config.temperature
               },
+              // Provider-specific options (caching, reasoning, etc.)
+              ...providerOptions !== void 0 && {
+                providerOptions
+              },
               // Log stream-level errors (tool errors, API errors during streaming)
               onError: (error) => {
                 this.logger.error("Stream error", { error });

package/dist/llm/executor/turn-executor.d.ts

@@ -51,6 +51,7 @@ export declare class TurnExecutor {
         maxOutputTokens?: number | undefined;
         temperature?: number | undefined;
         baseURL?: string | undefined;
+        reasoningEffort?: 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | undefined;
     }, llmContext: LLMContext, logger: IDextoLogger, messageQueue: MessageQueueService, modelLimits?: ModelLimits | undefined, externalSignal?: AbortSignal | undefined, compactionStrategy?: ICompactionStrategy | null);
     /**
      * Get StreamProcessor config from TurnExecutor state.

package/dist/llm/executor/turn-executor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"turn-executor.d.ts","sourceRoot":"","sources":["../../../src/llm/executor/turn-executor.ts"],"names":[],"mappings":"AAAA,OAAO,EACH,aAAa,EAMb,KAAK,YAAY,EAEpB,MAAM,IAAI,CAAC;AAEZ,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAE1D,OAAO,EAAE,WAAW,EAAE,MAAM,6BAA6B,CAAC;AAG1D,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAE5C,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAE7D,OAAO,KAAK,EAAE,eAAe,EAAmB,MAAM,uBAAuB,CAAC;AAC9E,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAChE,OAAO,EAAE,yBAAyB,EAAE,MAAM,6BAA6B,CAAC;AACxE,OAAO,EAAE,UAAU,EAAoB,MAAM,aAAa,CAAC;AAC3D,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,gCAAgC,CAAC;AAQ1E,OAAO,EAAc,KAAK,WAAW,EAAE,MAAM,sCAAsC,CAAC;AAEpF,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,mCAAmC,CAAC;AAc7E;;;;;;;;;;;GAWG;AACH,qBAAa,YAAY;IAkBjB,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,WAAW;IACnB,OAAO,CAAC,cAAc;IACtB,OAAO,CAAC,QAAQ;IAChB,OAAO,CAAC,eAAe;IACvB,OAAO,CAAC,SAAS;IACjB,OAAO,CAAC,MAAM;IAMd,OAAO,CAAC,UAAU;IAElB,OAAO,CAAC,YAAY;IACpB,OAAO,CAAC,WAAW,CAAC;IACpB,OAAO,CAAC,cAAc,CAAC;IAjC3B,OAAO,CAAC,MAAM,CAAe;IAC7B;;;OAGG;IACH,OAAO,CAAC,mBAAmB,CAAkB;IAC7C,OAAO,CAAC,kBAAkB,CAAoC;IAC9D;;;OAGG;IACH,OAAO,CAAC,gBAAgB,CAGpB;gBAGQ,KAAK,EAAE,aAAa,EACpB,WAAW,EAAE,WAAW,EACxB,cAAc,EAAE,cAAc,CAAC,YAAY,CAAC,EAC5C,QAAQ,EAAE,eAAe,EACzB,eAAe,EAAE,eAAe,EAChC,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE;QACZ,QAAQ,EAAE,MAAM,CAAC;QACjB,eAAe,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACrC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;KAChC,EACO,UAAU,EAAE,UAAU,EAC9B,MAAM,EAAE,YAAY,EACZ,YAAY,EAAE,mBAAmB,EACjC,WAAW,CAAC,EAAE,WAAW,YAAA,EACzB,cAAc,CAAC,EAAE,WAAW,YAAA,EACpC,kBAAkB,CAAC,EAAE,mBAAmB,GAAG,IAAI;IAqBnD;;OAEG;IACH,OAAO,CAAC,wBAAwB;IAOhC;;;;;;OAMG;IACG,OAAO,CACT,kBAAkB,EAAE,yBAAyB,EAC7C,SAAS,GAAE,OAAc,GAC1B,OAAO,CAAC,cAAc,CAAC;IA0M1B;;;OAGG;IACH,KAAK,IAAI,IAAI;IAIb;;;OAGG;YACW,oBAAoB;IAmBlC;;;;;;OAMG;YACW,mBAAmB;IA0EjC;;;;;;;OAOG;YACW,WAAW;IA0GzB;;;;;OAKG;IACH,OAAO,CAAC,sBAAsB;IAuF9B;;OAEG;IACH,OAAO,CAAC,gBAAgB;IAyBxB;;OAEG;IACH,OAAO,CAAC,eAAe;IAavB;;OAEG;IACH,OAAO,CAAC,oBAAoB;IAS5B;;OAEG;IACH,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,aAAa,CAAU;IAC/C,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,aAAa,CAAU;IAE/C;;;;;;;;;;;OAWG;YACW,mBAAmB;IAkDjC;;;;;;OAMG;IACH,OAAO,CAAC,kBAAkB;IAgB1B;;;OAGG;IACH,OAAO,CAAC,OAAO;IAYf;;OAEG;IACH,OAAO,CAAC,sBAAsB;IAO9B;;;;;;;;OAQG;YACW,QAAQ;IA8CtB;;OAEG;IACH,OAAO,CAAC,sBAAsB;IAoB9B;;OAEG;IACH,OAAO,CAAC,gBAAgB;CA0D3B"}
+{"version":3,"file":"turn-executor.d.ts","sourceRoot":"","sources":["../../../src/llm/executor/turn-executor.ts"],"names":[],"mappings":"AAAA,OAAO,EACH,aAAa,EAMb,KAAK,YAAY,EAEpB,MAAM,IAAI,CAAC;AAEZ,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAE1D,OAAO,EAAE,WAAW,EAAE,MAAM,6BAA6B,CAAC;AAG1D,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAG5C,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAE7D,OAAO,KAAK,EAAE,eAAe,EAAmB,MAAM,uBAAuB,CAAC;AAC9E,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAChE,OAAO,EAAE,yBAAyB,EAAE,MAAM,6BAA6B,CAAC;AACxE,OAAO,EAAE,UAAU,EAAoB,MAAM,aAAa,CAAC;AAC3D,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,gCAAgC,CAAC;AAQ1E,OAAO,EAAc,KAAK,WAAW,EAAE,MAAM,sCAAsC,CAAC;AAEpF,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,mCAAmC,CAAC;AAc7E;;;;;;;;;;;GAWG;AACH,qBAAa,YAAY;IAkBjB,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,WAAW;IACnB,OAAO,CAAC,cAAc;IACtB,OAAO,CAAC,QAAQ;IAChB,OAAO,CAAC,eAAe;IACvB,OAAO,CAAC,SAAS;IACjB,OAAO,CAAC,MAAM;IAQd,OAAO,CAAC,UAAU;IAElB,OAAO,CAAC,YAAY;IACpB,OAAO,CAAC,WAAW,CAAC;IACpB,OAAO,CAAC,cAAc,CAAC;IAnC3B,OAAO,CAAC,MAAM,CAAe;IAC7B;;;OAGG;IACH,OAAO,CAAC,mBAAmB,CAAkB;IAC7C,OAAO,CAAC,kBAAkB,CAAoC;IAC9D;;;OAGG;IACH,OAAO,CAAC,gBAAgB,CAGpB;gBAGQ,KAAK,EAAE,aAAa,EACpB,WAAW,EAAE,WAAW,EACxB,cAAc,EAAE,cAAc,CAAC,YAAY,CAAC,EAC5C,QAAQ,EAAE,eAAe,EACzB,eAAe,EAAE,eAAe,EAChC,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE;QACZ,QAAQ,EAAE,MAAM,CAAC;QACjB,eAAe,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACrC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QAE7B,eAAe,CAAC,EAAE,MAAM,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,SAAS,CAAC;KAC1F,EACO,UAAU,EAAE,UAAU,EAC9B,MAAM,EAAE,YAAY,EACZ,YAAY,EAAE,mBAAmB,EACjC,WAAW,CAAC,EAAE,WAAW,YAAA,EACzB,cAAc,CAAC,EAAE,WAAW,YAAA,EACpC,kBAAkB,CAAC,EAAE,mBAAmB,GAAG,IAAI;IAqBnD;;OAEG;IACH,OAAO,CAAC,wBAAwB;IAOhC;;;;;;OAMG;IACG,OAAO,CACT,kBAAkB,EAAE,yBAAyB,EAC7C,SAAS,GAAE,OAAc,GAC1B,OAAO,CAAC,cAAc,CAAC;IAsN1B;;;OAGG;IACH,KAAK,IAAI,IAAI;IAIb;;;OAGG;YACW,oBAAoB;IAmBlC;;;;;;OAMG;YACW,mBAAmB;IA0EjC;;;;;;;OAOG;YACW,WAAW;IA0GzB;;;;;OAKG;IACH,OAAO,CAAC,sBAAsB;IAuF9B;;OAEG;IACH,OAAO,CAAC,gBAAgB;IAyBxB;;OAEG;IACH,OAAO,CAAC,eAAe;IAavB;;OAEG;IACH,OAAO,CAAC,oBAAoB;IAS5B;;OAEG;IACH,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,aAAa,CAAU;IAC/C,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,aAAa,CAAU;IAE/C;;;;;;;;;;;OAWG;YACW,mBAAmB;IAkDjC;;;;;;OAMG;IACH,OAAO,CAAC,kBAAkB;IAgB1B;;;OAGG;IACH,OAAO,CAAC,OAAO;IAYf;;OAEG;IACH,OAAO,CAAC,sBAAsB;IAO9B;;;;;;;;OAQG;YACW,QAAQ;IA8CtB;;OAEG;IACH,OAAO,CAAC,sBAAsB;IAoB9B;;OAEG;IACH,OAAO,CAAC,gBAAgB;CA0D3B"}

package/dist/llm/executor/turn-executor.js

@@ -11,6 +11,7 @@ import {
 } from "ai";
 import { trace } from "@opentelemetry/api";
 import { StreamProcessor } from "./stream-processor.js";
+import { buildProviderOptions } from "./provider-options.js";
 import { DextoLogComponent } from "../../logger/v2/types.js";
 import { defer } from "../../utils/defer.js";
 import { DextoRuntimeError } from "../../errors/DextoRuntimeError.js";
@@ -135,6 +136,11 @@ class TurnExecutor {
             streaming,
             this.approvalMetadata
           );
+          const providerOptions = buildProviderOptions({
+            provider: this.llmContext.provider,
+            model: this.llmContext.model,
+            reasoningEffort: this.config.reasoningEffort
+          });
           const result = await streamProcessor.process(
             () => streamText({
               model: this.model,
@@ -148,6 +154,10 @@ class TurnExecutor {
               ...this.config.temperature !== void 0 && {
                 temperature: this.config.temperature
               },
+              // Provider-specific options (caching, reasoning, etc.)
+              ...providerOptions !== void 0 && {
+                providerOptions
+              },
               // Log stream-level errors (tool errors, API errors during streaming)
               onError: (error) => {
                 this.logger.error("Stream error", { error });

package/dist/llm/formatters/vercel.cjs

@@ -59,9 +59,17 @@ class VercelMessageFormatter {
       filteredHistory = [...history];
     }
     if (systemPrompt) {
+      const modelLower = context.model.toLowerCase();
+      const isClaudeModel = modelLower.includes("claude");
+      const isAnthropicProvider = context.provider === "anthropic" || context.provider === "bedrock" && isClaudeModel || context.provider === "vertex" && isClaudeModel;
       formatted.push({
         role: "system",
-        content: systemPrompt
+        content: systemPrompt,
+        ...isAnthropicProvider && {
+          providerOptions: {
+            anthropic: { cacheControl: { type: "ephemeral" } }
+          }
+        }
       });
     }
     const pendingToolCalls = /* @__PURE__ */ new Map();

package/dist/llm/formatters/vercel.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"vercel.d.ts","sourceRoot":"","sources":["../../../src/llm/formatters/vercel.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAiD,MAAM,IAAI,CAAC;AACtF,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,KAAK,EAAE,eAAe,EAAiC,MAAM,wBAAwB,CAAC;AAE7F,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAmB7D;;;;;;;;;;GAUG;AACH,qBAAa,sBAAsB;IAC/B,OAAO,CAAC,MAAM,CAAe;gBAEjB,MAAM,EAAE,YAAY;IAGhC;;;;;;OAMG;IACH,MAAM,CACF,OAAO,EAAE,QAAQ,CAAC,eAAe,EAAE,CAAC,EACpC,OAAO,EAAE,UAAU,EACnB,YAAY,EAAE,MAAM,GAAG,IAAI,GAC5B,YAAY,EAAE;IAwIjB;;;;;;OAMG;IACH,kBAAkB,IAAI,IAAI;IAK1B,OAAO,CAAC,sBAAsB;IA8E9B,OAAO,CAAC,iBAAiB;CAoE5B"}
+{"version":3,"file":"vercel.d.ts","sourceRoot":"","sources":["../../../src/llm/formatters/vercel.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAiD,MAAM,IAAI,CAAC;AACtF,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,KAAK,EAAE,eAAe,EAAiC,MAAM,wBAAwB,CAAC;AAE7F,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAmB7D;;;;;;;;;;GAUG;AACH,qBAAa,sBAAsB;IAC/B,OAAO,CAAC,MAAM,CAAe;gBAEjB,MAAM,EAAE,YAAY;IAGhC;;;;;;OAMG;IACH,MAAM,CACF,OAAO,EAAE,QAAQ,CAAC,eAAe,EAAE,CAAC,EACpC,OAAO,EAAE,UAAU,EACnB,YAAY,EAAE,MAAM,GAAG,IAAI,GAC5B,YAAY,EAAE;IAsJjB;;;;;;OAMG;IACH,kBAAkB,IAAI,IAAI;IAK1B,OAAO,CAAC,sBAAsB;IA8E9B,OAAO,CAAC,iBAAiB;CAoE5B"}

package/dist/llm/formatters/vercel.js

@@ -37,9 +37,17 @@ class VercelMessageFormatter {
       filteredHistory = [...history];
     }
     if (systemPrompt) {
+      const modelLower = context.model.toLowerCase();
+      const isClaudeModel = modelLower.includes("claude");
+      const isAnthropicProvider = context.provider === "anthropic" || context.provider === "bedrock" && isClaudeModel || context.provider === "vertex" && isClaudeModel;
       formatted.push({
         role: "system",
-        content: systemPrompt
+        content: systemPrompt,
+        ...isAnthropicProvider && {
+          providerOptions: {
+            anthropic: { cacheControl: { type: "ephemeral" } }
+          }
+        }
       });
     }
     const pendingToolCalls = /* @__PURE__ */ new Map();

package/dist/llm/registry.cjs

@@ -34,6 +34,7 @@ __export(registry_exports, {
   getSupportedFileTypesForModel: () => getSupportedFileTypesForModel,
   getSupportedModels: () => getSupportedModels,
   getSupportedProviders: () => getSupportedProviders,
+  isReasoningCapableModel: () => isReasoningCapableModel,
   isValidProviderModel: () => isValidProviderModel,
   modelSupportsFileType: () => modelSupportsFileType,
   requiresApiKey: () => requiresApiKey,
@@ -74,6 +75,60 @@ const DEFAULT_MAX_INPUT_TOKENS = 128e3;
 const LLM_REGISTRY = {
   openai: {
     models: [
+      // GPT-5.2 series (latest, released Dec 2025)
+      {
+        name: "gpt-5.2-chat-latest",
+        displayName: "GPT-5.2 Instant",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2",
+        displayName: "GPT-5.2 Thinking",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2-pro",
+        displayName: "GPT-5.2 Pro",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 21,
+          outputPerM: 168,
+          cacheReadPerM: 2.1,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2-codex",
+        displayName: "GPT-5.2 Codex",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // GPT-5.1 series
       {
         name: "gpt-5.1-chat-latest",
         displayName: "GPT-5.1 Instant",
@@ -1525,6 +1580,19 @@ function getModelDisplayName(model, provider) {
   const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   return modelInfo?.displayName ?? model;
 }
+function isReasoningCapableModel(model, provider) {
+  const modelLower = model.toLowerCase();
+  if (modelLower.includes("codex")) {
+    return true;
+  }
+  if (modelLower.startsWith("o1") || modelLower.startsWith("o3") || modelLower.startsWith("o4")) {
+    return true;
+  }
+  if (modelLower.includes("gpt-5") || modelLower.includes("gpt-5.1") || modelLower.includes("gpt-5.2")) {
+    return true;
+  }
+  return false;
+}
 function calculateCost(usage, pricing) {
   const inputCost = (usage.inputTokens ?? 0) * pricing.inputPerM / 1e6;
   const outputCost = (usage.outputTokens ?? 0) * pricing.outputPerM / 1e6;
@@ -1551,6 +1619,7 @@ function calculateCost(usage, pricing) {
   getSupportedFileTypesForModel,
   getSupportedModels,
   getSupportedProviders,
+  isReasoningCapableModel,
   isValidProviderModel,
   modelSupportsFileType,
   requiresApiKey,

package/dist/llm/registry.d.ts

@@ -205,6 +205,15 @@ export declare function getModelPricing(provider: LLMProvider, model: string): M
  * Gets the display name for a model, falling back to the model ID if not found.
  */
 export declare function getModelDisplayName(model: string, provider?: LLMProvider): string;
+/**
+ * Checks if a model supports configurable reasoning effort.
+ * Currently only OpenAI reasoning models (o1, o3, codex, gpt-5.x) support this.
+ *
+ * @param model The model name to check.
+ * @param provider Optional provider for context (defaults to detecting from model name).
+ * @returns True if the model supports reasoning effort configuration.
+ */
+export declare function isReasoningCapableModel(model: string, provider?: LLMProvider): boolean;
 /**
  * Calculates the cost for a given token usage based on model pricing.
  *

package/dist/llm/registry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../../src/llm/registry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAIzC,OAAO,EAEH,KAAK,WAAW,EAChB,KAAK,iBAAiB,EACtB,KAAK,UAAU,EAClB,MAAM,YAAY,CAAC;AACpB,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AAG1D;;;GAGG;AACH,MAAM,WAAW,YAAY;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,EAAE,KAAK,CAAC;IACjB,IAAI,CAAC,EAAE,oBAAoB,CAAC;CAC/B;AAED,MAAM,WAAW,SAAS;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,kBAAkB,EAAE,iBAAiB,EAAE,CAAC;IACxC,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,OAAO,CAAC,EAAE,YAAY,CAAC;CAE1B;AAOD,eAAO,MAAM,sBAAsB,EAAE,MAAM,CAAC,MAAM,EAAE,iBAAiB,CAiBpE,CAAC;AAGF,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C;AAED,MAAM,WAAW,YAAY;IACzB,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,cAAc,EAAE,MAAM,GAAG,UAAU,GAAG,UAAU,CAAC;IACjD,kBAAkB,EAAE,iBAAiB,EAAE,CAAC;IACxC,oBAAoB,CAAC,EAAE,OAAO,CAAC;CAElC;AAED,sEAAsE;AACtE,eAAO,MAAM,wBAAwB,SAAS,CAAC;AAI/C;;;;;;;;GAQG;AACH,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,WAAW,EAAE,YAAY,CA2pC1D,CAAC;AAEF;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAQ9D;AAED;;;;GAIG;AACH,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,WAAW,GAAG,MAAM,GAAG,IAAI,CAG/E;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,IAAI,WAAW,EAAE,CAErD;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,WAAW,GAAG,MAAM,EAAE,CAGlE;AAED;;;;;;;GAOG;AACH,wBAAgB,yBAAyB,CACrC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,MAAM,CAAC,EAAE,YAAY,GACtB,MAAM,CAeR;AAED;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAIlF;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,WAAW,CAS/D;AAED;;GAEG;AACH,wBAAgB,qBAAqB,IAAI,MAAM,EAAE,CAEhD;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAGnE;AAmBD;;;;;;;;;GASG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAE7D;AAED;;;;;;GAMG;AACH,wBAAgB,6BAA6B,CACzC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,GACd,iBAAiB,EAAE,CAkBrB;AAED;;;;;;GAMG;AACH,wBAAgB,qBAAqB,CACjC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,iBAAiB,GAC5B,OAAO,CAGT;AAED;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACpC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM,GACjB;IACC,WAAW,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,iBAAiB,CAAC;IAC7B,KAAK,CAAC,EAAE,MAAM,CAAC;CAClB,CAkCA;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,0BAA0B,CAAC,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,GAAG,MAAM,CAqH1F;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAW9F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,WAAW,GAAG,MAAM,CAkBjF;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,UAAU,EAAE,OAAO,EAAE,YAAY,GAAG,MAAM,CAU9E"}
+{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../../src/llm/registry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAIzC,OAAO,EAEH,KAAK,WAAW,EAChB,KAAK,iBAAiB,EACtB,KAAK,UAAU,EAClB,MAAM,YAAY,CAAC;AACpB,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AAG1D;;;GAGG;AACH,MAAM,WAAW,YAAY;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,EAAE,KAAK,CAAC;IACjB,IAAI,CAAC,EAAE,oBAAoB,CAAC;CAC/B;AAED,MAAM,WAAW,SAAS;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,kBAAkB,EAAE,iBAAiB,EAAE,CAAC;IACxC,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,OAAO,CAAC,EAAE,YAAY,CAAC;CAE1B;AAOD,eAAO,MAAM,sBAAsB,EAAE,MAAM,CAAC,MAAM,EAAE,iBAAiB,CAiBpE,CAAC;AAGF,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C;AAED,MAAM,WAAW,YAAY;IACzB,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,cAAc,EAAE,MAAM,GAAG,UAAU,GAAG,UAAU,CAAC;IACjD,kBAAkB,EAAE,iBAAiB,EAAE,CAAC;IACxC,oBAAoB,CAAC,EAAE,OAAO,CAAC;CAElC;AAED,sEAAsE;AACtE,eAAO,MAAM,wBAAwB,SAAS,CAAC;AAI/C;;;;;;;;GAQG;AACH,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,WAAW,EAAE,YAAY,CAitC1D,CAAC;AAEF;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAQ9D;AAED;;;;GAIG;AACH,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,WAAW,GAAG,MAAM,GAAG,IAAI,CAG/E;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,IAAI,WAAW,EAAE,CAErD;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,WAAW,GAAG,MAAM,EAAE,CAGlE;AAED;;;;;;;GAOG;AACH,wBAAgB,yBAAyB,CACrC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,MAAM,CAAC,EAAE,YAAY,GACtB,MAAM,CAeR;AAED;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAIlF;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,WAAW,CAS/D;AAED;;GAEG;AACH,wBAAgB,qBAAqB,IAAI,MAAM,EAAE,CAEhD;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAG9D;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAGnE;AAmBD;;;;;;;;;GASG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,WAAW,GAAG,OAAO,CAE7D;AAED;;;;;;GAMG;AACH,wBAAgB,6BAA6B,CACzC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,GACd,iBAAiB,EAAE,CAkBrB;AAED;;;;;;GAMG;AACH,wBAAgB,qBAAqB,CACjC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,iBAAiB,GAC5B,OAAO,CAGT;AAED;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACpC,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM,GACjB;IACC,WAAW,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,iBAAiB,CAAC;IAC7B,KAAK,CAAC,EAAE,MAAM,CAAC;CAClB,CAkCA;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,0BAA0B,CAAC,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,GAAG,MAAM,CAqH1F;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAW9F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,WAAW,GAAG,MAAM,CAkBjF;AAED;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,WAAW,GAAG,OAAO,CAuBtF;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,UAAU,EAAE,OAAO,EAAE,YAAY,GAAG,MAAM,CAU9E"}

package/dist/llm/registry.js

@@ -31,6 +31,60 @@ const DEFAULT_MAX_INPUT_TOKENS = 128e3;
 const LLM_REGISTRY = {
   openai: {
     models: [
+      // GPT-5.2 series (latest, released Dec 2025)
+      {
+        name: "gpt-5.2-chat-latest",
+        displayName: "GPT-5.2 Instant",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2",
+        displayName: "GPT-5.2 Thinking",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2-pro",
+        displayName: "GPT-5.2 Pro",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 21,
+          outputPerM: 168,
+          cacheReadPerM: 2.1,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gpt-5.2-codex",
+        displayName: "GPT-5.2 Codex",
+        maxInputTokens: 4e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1.75,
+          outputPerM: 14,
+          cacheReadPerM: 0.175,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // GPT-5.1 series
       {
         name: "gpt-5.1-chat-latest",
         displayName: "GPT-5.1 Instant",
@@ -1482,6 +1536,19 @@ function getModelDisplayName(model, provider) {
   const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   return modelInfo?.displayName ?? model;
 }
+function isReasoningCapableModel(model, provider) {
+  const modelLower = model.toLowerCase();
+  if (modelLower.includes("codex")) {
+    return true;
+  }
+  if (modelLower.startsWith("o1") || modelLower.startsWith("o3") || modelLower.startsWith("o4")) {
+    return true;
+  }
+  if (modelLower.includes("gpt-5") || modelLower.includes("gpt-5.1") || modelLower.includes("gpt-5.2")) {
+    return true;
+  }
+  return false;
+}
 function calculateCost(usage, pricing) {
   const inputCost = (usage.inputTokens ?? 0) * pricing.inputPerM / 1e6;
   const outputCost = (usage.outputTokens ?? 0) * pricing.outputPerM / 1e6;
@@ -1507,6 +1574,7 @@ export {
   getSupportedFileTypesForModel,
   getSupportedModels,
   getSupportedProviders,
+  isReasoningCapableModel,
   isValidProviderModel,
   modelSupportsFileType,
   requiresApiKey,