@yourgpt/llm-sdk 2.5.0 → 2.5.1-beta.0

package/dist/adapters/index.d.mts

@@ -1,7 +1,7 @@
-import { L as LLMAdapter, W as WebSearchConfig, C as ChatCompletionRequest, S as StreamEvent, o as CompletionResult } from '../base-D-U61JaB.mjs';
-export { e as AdapterFactory, B as AnthropicContentBlock, E as OpenAIContentBlock, y as attachmentToAnthropicDocument, x as attachmentToAnthropicImage, z as attachmentToOpenAIImage, p as formatMessages, r as formatMessagesForAnthropic, s as formatMessagesForOpenAI, q as formatTools, v as hasImageAttachments, w as hasMediaAttachments, t as messageToAnthropicContent, u as messageToOpenAIContent } from '../base-D-U61JaB.mjs';
-import { d as OllamaModelOptions } from '../types-DRqxMIjF.mjs';
-import '../types-CR8mi9I0.mjs';
+import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-tNgbBaSo.mjs';
+export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-tNgbBaSo.mjs';
+import { W as WebSearchConfig, g as StreamEvent } from '../types-BkQCSiIt.mjs';
+import { d as OllamaModelOptions } from '../types-BSSiJW2o.mjs';
 import 'zod';
 
 /**

package/dist/adapters/index.d.ts

@@ -1,7 +1,7 @@
-import { L as LLMAdapter, W as WebSearchConfig, C as ChatCompletionRequest, S as StreamEvent, o as CompletionResult } from '../base-iGi9Va6Z.js';
-export { e as AdapterFactory, B as AnthropicContentBlock, E as OpenAIContentBlock, y as attachmentToAnthropicDocument, x as attachmentToAnthropicImage, z as attachmentToOpenAIImage, p as formatMessages, r as formatMessagesForAnthropic, s as formatMessagesForOpenAI, q as formatTools, v as hasImageAttachments, w as hasMediaAttachments, t as messageToAnthropicContent, u as messageToOpenAIContent } from '../base-iGi9Va6Z.js';
-import { d as OllamaModelOptions } from '../types-BctsnC3g.js';
-import '../types-CR8mi9I0.js';
+import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-C58Dsr9p.js';
+export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-C58Dsr9p.js';
+import { W as WebSearchConfig, g as StreamEvent } from '../types-BkQCSiIt.js';
+import { d as OllamaModelOptions } from '../types-BQ31QIsA.js';
 import 'zod';
 
 /**

package/dist/adapters/index.js

@@ -114,6 +114,109 @@ function normalizeObjectJsonSchema(schema) {
   }
   return normalized;
 }
+function isOpenAIReasoningModel(modelId) {
+  if (!modelId) return false;
+  return /^(o1|o3|o4|gpt-5)/i.test(modelId);
+}
+function buildOpenAITokenParams(modelId, maxTokens, temperature) {
+  if (isOpenAIReasoningModel(modelId)) {
+    return { max_completion_tokens: maxTokens };
+  }
+  return { max_tokens: maxTokens, temperature };
+}
+function stripSchemaKeys(schema, keysToDrop, options = {}) {
+  if (Array.isArray(schema)) {
+    return schema.map((item) => stripSchemaKeys(item, keysToDrop, options));
+  }
+  if (!schema || typeof schema !== "object") return schema;
+  const out = {};
+  for (const [key, value] of Object.entries(
+    schema
+  )) {
+    if (keysToDrop.has(key)) continue;
+    const renamed = options.renameKeys?.[key] ?? key;
+    out[renamed] = stripSchemaKeys(value, keysToDrop, options);
+  }
+  if (options.forceAdditionalPropertiesFalse && out.type === "object") {
+    out.additionalProperties = false;
+  }
+  return out;
+}
+function toOpenAIResponseFormat(rf) {
+  if (!rf) return void 0;
+  if (rf.type === "json_object") return { type: "json_object" };
+  return {
+    type: "json_schema",
+    json_schema: {
+      name: rf.json_schema.name,
+      schema: normalizeObjectJsonSchema(rf.json_schema.schema),
+      strict: rf.json_schema.strict ?? true
+    }
+  };
+}
+function toOpenAIResponsesTextFormat(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  return {
+    type: "json_schema",
+    name: rf.json_schema.name,
+    schema: normalizeObjectJsonSchema(rf.json_schema.schema),
+    strict: rf.json_schema.strict ?? true
+  };
+}
+var ANTHROPIC_UNSUPPORTED_KEYS = /* @__PURE__ */ new Set([
+  "minimum",
+  "maximum",
+  "exclusiveMinimum",
+  "exclusiveMaximum",
+  "multipleOf",
+  "minLength",
+  "maxLength",
+  "minItems",
+  "maxItems",
+  "minProperties",
+  "maxProperties",
+  "pattern",
+  "$schema"
+]);
+function toAnthropicOutputConfig(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  const schema = stripSchemaKeys(
+    rf.json_schema.schema,
+    ANTHROPIC_UNSUPPORTED_KEYS,
+    {
+      forceAdditionalPropertiesFalse: true,
+      renameKeys: { oneOf: "anyOf" }
+    }
+  );
+  return {
+    format: {
+      type: "json_schema",
+      schema
+    }
+  };
+}
+var GEMINI_UNSUPPORTED_KEYS = /* @__PURE__ */ new Set([
+  "oneOf",
+  "anyOf",
+  "$ref",
+  "$defs",
+  "definitions",
+  "pattern",
+  "$schema",
+  "additionalProperties"
+]);
+function toGeminiSchema(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  return stripSchemaKeys(
+    rf.json_schema.schema,
+    GEMINI_UNSUPPORTED_KEYS
+  );
+}
+function toOllamaFormat(rf) {
+  if (!rf) return void 0;
+  if (rf.type === "json_object") return "json";
+  return rf.json_schema.schema;
+}
 function formatTools(actions) {
   return actions.map((action) => ({
     type: "function",
@@ -484,6 +587,9 @@ var OpenAIAdapter = class _OpenAIAdapter {
   async completeWithResponses(request) {
     const client = await this.getClient();
     const openaiToolOptions = request.providerToolOptions?.openai;
+    const responsesTextFormat = toOpenAIResponsesTextFormat(
+      request.config?.responseFormat
+    );
     const payload = {
       model: request.config?.model || this.model,
       instructions: request.systemPrompt,
@@ -493,6 +599,7 @@ var OpenAIAdapter = class _OpenAIAdapter {
       parallel_tool_calls: openaiToolOptions?.parallelToolCalls,
       temperature: request.config?.temperature ?? this.config.temperature,
       max_output_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      ...responsesTextFormat ? { text: { format: responsesTextFormat } } : {},
       stream: false
     };
     logProviderPayload("openai", "request payload", payload, request.debug);
@@ -614,14 +721,19 @@ var OpenAIAdapter = class _OpenAIAdapter {
           name: openaiToolOptions.toolChoice.name
         }
       } : openaiToolOptions?.toolChoice;
+      const modelIdForPayload = request.config?.model || this.model;
       const payload = {
-        model: request.config?.model || this.model,
+        model: modelIdForPayload,
         messages,
         tools: tools.length > 0 ? tools : void 0,
         tool_choice: tools.length > 0 ? toolChoice : void 0,
         parallel_tool_calls: tools.length > 0 ? openaiToolOptions?.parallelToolCalls : void 0,
-        temperature: request.config?.temperature ?? this.config.temperature,
-        max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...buildOpenAITokenParams(
+          modelIdForPayload,
+          request.config?.maxTokens ?? this.config.maxTokens,
+          request.config?.temperature ?? this.config.temperature
+        ),
+        response_format: toOpenAIResponseFormat(request.config?.responseFormat),
         stream: true,
         stream_options: { include_usage: true }
       };
@@ -763,14 +875,19 @@ var OpenAIAdapter = class _OpenAIAdapter {
         name: openaiToolOptions.toolChoice.name
       }
     } : openaiToolOptions?.toolChoice;
+    const modelIdForCompletePayload = request.config?.model || this.model;
     const payload = {
-      model: request.config?.model || this.model,
+      model: modelIdForCompletePayload,
       messages,
       tools: tools.length > 0 ? tools : void 0,
       tool_choice: tools.length > 0 ? toolChoice : void 0,
       parallel_tool_calls: tools.length > 0 ? openaiToolOptions?.parallelToolCalls : void 0,
-      temperature: request.config?.temperature ?? this.config.temperature,
-      max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      ...buildOpenAITokenParams(
+        modelIdForCompletePayload,
+        request.config?.maxTokens ?? this.config.maxTokens,
+        request.config?.temperature ?? this.config.temperature
+      ),
+      response_format: toOpenAIResponseFormat(request.config?.responseFormat),
       stream: false
     };
     logProviderPayload("openai", "request payload", payload, request.debug);
@@ -1046,7 +1163,9 @@ var AnthropicAdapter = class {
    * Build common request options for both streaming and non-streaming
    */
   buildRequestOptions(request) {
-    const systemMessage = request.systemPrompt || "";
+    const responseFormat = request.config?.responseFormat;
+    const jsonObjectSuffix = responseFormat?.type === "json_object" ? "\n\nRespond with a single JSON object and no other text." : "";
+    const systemMessage = (request.systemPrompt || "") + jsonObjectSuffix;
     let messages;
     if (request.rawMessages && request.rawMessages.length > 0) {
       messages = this.convertToAnthropicMessages(request.rawMessages);
@@ -1127,6 +1246,10 @@ var AnthropicAdapter = class {
     if (serverToolConfiguration) {
       options.server_tool_configuration = serverToolConfiguration;
     }
+    const outputConfig = toAnthropicOutputConfig(responseFormat);
+    if (outputConfig) {
+      options.output_config = outputConfig;
+    }
     if (this.config.thinking?.type === "enabled") {
       options.thinking = {
         type: "enabled",
@@ -1487,12 +1610,14 @@ var OllamaAdapter = class {
       if (this.config.options) {
         Object.assign(ollamaOptions, this.config.options);
       }
+      const ollamaFormat = toOllamaFormat(request.config?.responseFormat);
       const payload = {
         model: request.config?.model || this.model,
         messages,
         tools,
         stream: true,
-        options: ollamaOptions
+        options: ollamaOptions,
+        ...ollamaFormat !== void 0 ? { format: ollamaFormat } : {}
       };
       logProviderPayload("ollama", "request payload", payload, request.debug);
       const response = await fetch(`${this.baseUrl}/api/chat`, {
@@ -1783,6 +1908,12 @@ var GoogleAdapter = class {
     }
     const messageId = generateMessageId();
     yield { type: "message:start", id: messageId };
+    const responseFormat = request.config?.responseFormat;
+    const geminiSchema = toGeminiSchema(responseFormat);
+    const responseFormatGenConfig = responseFormat ? {
+      responseMimeType: "application/json",
+      ...geminiSchema ? { responseJsonSchema: geminiSchema } : {}
+    } : {};
     try {
       logProviderPayload(
         "google",
@@ -1794,7 +1925,8 @@ var GoogleAdapter = class {
           tools: toolsArray.length > 0 ? toolsArray : void 0,
           generationConfig: {
             temperature: request.config?.temperature ?? this.config.temperature,
-            maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+            maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+            ...responseFormatGenConfig
           },
           messageParts: mergedContents[mergedContents.length - 1]?.parts
         },
@@ -1807,7 +1939,8 @@ var GoogleAdapter = class {
         tools: toolsArray.length > 0 ? toolsArray : void 0,
         generationConfig: {
           temperature: request.config?.temperature ?? this.config.temperature,
-          maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+          maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+          ...responseFormatGenConfig
         }
       });
       const lastMessage = mergedContents[mergedContents.length - 1];
@@ -1974,6 +2107,12 @@ var GoogleAdapter = class {
       }
     }
     const tools = formatToolsForGemini(request.actions);
+    const responseFormat = request.config?.responseFormat;
+    const geminiSchema = toGeminiSchema(responseFormat);
+    const responseFormatGenConfig = responseFormat ? {
+      responseMimeType: "application/json",
+      ...geminiSchema ? { responseJsonSchema: geminiSchema } : {}
+    } : {};
     const payload = {
       model: modelId,
       history: mergedContents.slice(0, -1),
@@ -1981,7 +2120,8 @@ var GoogleAdapter = class {
       tools: tools ? [tools] : void 0,
       generationConfig: {
         temperature: request.config?.temperature ?? this.config.temperature,
-        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...responseFormatGenConfig
       },
       messageParts: mergedContents[mergedContents.length - 1]?.parts
     };
@@ -1992,7 +2132,8 @@ var GoogleAdapter = class {
       tools: tools ? [tools] : void 0,
       generationConfig: {
         temperature: request.config?.temperature ?? this.config.temperature,
-        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...responseFormatGenConfig
       }
     });
     const lastMessage = mergedContents[mergedContents.length - 1];
@@ -2131,6 +2272,7 @@ var AzureAdapter = class {
         tools,
         temperature: request.config?.temperature ?? this.config.temperature,
         max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+        response_format: toOpenAIResponseFormat(request.config?.responseFormat),
         stream: true
       };
       logProviderPayload("azure", "request payload", payload, request.debug);
@@ -2230,7 +2372,8 @@ var AzureAdapter = class {
       messages,
       tools,
       temperature: request.config?.temperature ?? this.config.temperature,
-      max_tokens: request.config?.maxTokens ?? this.config.maxTokens
+      max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      response_format: toOpenAIResponseFormat(request.config?.responseFormat)
     };
     logProviderPayload("azure", "request payload", payload, request.debug);
     const response = await client.chat.completions.create(payload);

package/dist/adapters/index.mjs

@@ -112,6 +112,109 @@ function normalizeObjectJsonSchema(schema) {
   }
   return normalized;
 }
+function isOpenAIReasoningModel(modelId) {
+  if (!modelId) return false;
+  return /^(o1|o3|o4|gpt-5)/i.test(modelId);
+}
+function buildOpenAITokenParams(modelId, maxTokens, temperature) {
+  if (isOpenAIReasoningModel(modelId)) {
+    return { max_completion_tokens: maxTokens };
+  }
+  return { max_tokens: maxTokens, temperature };
+}
+function stripSchemaKeys(schema, keysToDrop, options = {}) {
+  if (Array.isArray(schema)) {
+    return schema.map((item) => stripSchemaKeys(item, keysToDrop, options));
+  }
+  if (!schema || typeof schema !== "object") return schema;
+  const out = {};
+  for (const [key, value] of Object.entries(
+    schema
+  )) {
+    if (keysToDrop.has(key)) continue;
+    const renamed = options.renameKeys?.[key] ?? key;
+    out[renamed] = stripSchemaKeys(value, keysToDrop, options);
+  }
+  if (options.forceAdditionalPropertiesFalse && out.type === "object") {
+    out.additionalProperties = false;
+  }
+  return out;
+}
+function toOpenAIResponseFormat(rf) {
+  if (!rf) return void 0;
+  if (rf.type === "json_object") return { type: "json_object" };
+  return {
+    type: "json_schema",
+    json_schema: {
+      name: rf.json_schema.name,
+      schema: normalizeObjectJsonSchema(rf.json_schema.schema),
+      strict: rf.json_schema.strict ?? true
+    }
+  };
+}
+function toOpenAIResponsesTextFormat(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  return {
+    type: "json_schema",
+    name: rf.json_schema.name,
+    schema: normalizeObjectJsonSchema(rf.json_schema.schema),
+    strict: rf.json_schema.strict ?? true
+  };
+}
+var ANTHROPIC_UNSUPPORTED_KEYS = /* @__PURE__ */ new Set([
+  "minimum",
+  "maximum",
+  "exclusiveMinimum",
+  "exclusiveMaximum",
+  "multipleOf",
+  "minLength",
+  "maxLength",
+  "minItems",
+  "maxItems",
+  "minProperties",
+  "maxProperties",
+  "pattern",
+  "$schema"
+]);
+function toAnthropicOutputConfig(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  const schema = stripSchemaKeys(
+    rf.json_schema.schema,
+    ANTHROPIC_UNSUPPORTED_KEYS,
+    {
+      forceAdditionalPropertiesFalse: true,
+      renameKeys: { oneOf: "anyOf" }
+    }
+  );
+  return {
+    format: {
+      type: "json_schema",
+      schema
+    }
+  };
+}
+var GEMINI_UNSUPPORTED_KEYS = /* @__PURE__ */ new Set([
+  "oneOf",
+  "anyOf",
+  "$ref",
+  "$defs",
+  "definitions",
+  "pattern",
+  "$schema",
+  "additionalProperties"
+]);
+function toGeminiSchema(rf) {
+  if (!rf || rf.type !== "json_schema") return void 0;
+  return stripSchemaKeys(
+    rf.json_schema.schema,
+    GEMINI_UNSUPPORTED_KEYS
+  );
+}
+function toOllamaFormat(rf) {
+  if (!rf) return void 0;
+  if (rf.type === "json_object") return "json";
+  return rf.json_schema.schema;
+}
 function formatTools(actions) {
   return actions.map((action) => ({
     type: "function",
@@ -482,6 +585,9 @@ var OpenAIAdapter = class _OpenAIAdapter {
   async completeWithResponses(request) {
     const client = await this.getClient();
     const openaiToolOptions = request.providerToolOptions?.openai;
+    const responsesTextFormat = toOpenAIResponsesTextFormat(
+      request.config?.responseFormat
+    );
     const payload = {
       model: request.config?.model || this.model,
       instructions: request.systemPrompt,
@@ -491,6 +597,7 @@ var OpenAIAdapter = class _OpenAIAdapter {
       parallel_tool_calls: openaiToolOptions?.parallelToolCalls,
       temperature: request.config?.temperature ?? this.config.temperature,
       max_output_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      ...responsesTextFormat ? { text: { format: responsesTextFormat } } : {},
       stream: false
     };
     logProviderPayload("openai", "request payload", payload, request.debug);
@@ -612,14 +719,19 @@ var OpenAIAdapter = class _OpenAIAdapter {
           name: openaiToolOptions.toolChoice.name
         }
       } : openaiToolOptions?.toolChoice;
+      const modelIdForPayload = request.config?.model || this.model;
       const payload = {
-        model: request.config?.model || this.model,
+        model: modelIdForPayload,
         messages,
         tools: tools.length > 0 ? tools : void 0,
         tool_choice: tools.length > 0 ? toolChoice : void 0,
         parallel_tool_calls: tools.length > 0 ? openaiToolOptions?.parallelToolCalls : void 0,
-        temperature: request.config?.temperature ?? this.config.temperature,
-        max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...buildOpenAITokenParams(
+          modelIdForPayload,
+          request.config?.maxTokens ?? this.config.maxTokens,
+          request.config?.temperature ?? this.config.temperature
+        ),
+        response_format: toOpenAIResponseFormat(request.config?.responseFormat),
         stream: true,
         stream_options: { include_usage: true }
       };
@@ -761,14 +873,19 @@ var OpenAIAdapter = class _OpenAIAdapter {
         name: openaiToolOptions.toolChoice.name
       }
     } : openaiToolOptions?.toolChoice;
+    const modelIdForCompletePayload = request.config?.model || this.model;
     const payload = {
-      model: request.config?.model || this.model,
+      model: modelIdForCompletePayload,
       messages,
       tools: tools.length > 0 ? tools : void 0,
       tool_choice: tools.length > 0 ? toolChoice : void 0,
       parallel_tool_calls: tools.length > 0 ? openaiToolOptions?.parallelToolCalls : void 0,
-      temperature: request.config?.temperature ?? this.config.temperature,
-      max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      ...buildOpenAITokenParams(
+        modelIdForCompletePayload,
+        request.config?.maxTokens ?? this.config.maxTokens,
+        request.config?.temperature ?? this.config.temperature
+      ),
+      response_format: toOpenAIResponseFormat(request.config?.responseFormat),
       stream: false
     };
     logProviderPayload("openai", "request payload", payload, request.debug);
@@ -1044,7 +1161,9 @@ var AnthropicAdapter = class {
    * Build common request options for both streaming and non-streaming
    */
   buildRequestOptions(request) {
-    const systemMessage = request.systemPrompt || "";
+    const responseFormat = request.config?.responseFormat;
+    const jsonObjectSuffix = responseFormat?.type === "json_object" ? "\n\nRespond with a single JSON object and no other text." : "";
+    const systemMessage = (request.systemPrompt || "") + jsonObjectSuffix;
     let messages;
     if (request.rawMessages && request.rawMessages.length > 0) {
       messages = this.convertToAnthropicMessages(request.rawMessages);
@@ -1125,6 +1244,10 @@ var AnthropicAdapter = class {
     if (serverToolConfiguration) {
       options.server_tool_configuration = serverToolConfiguration;
     }
+    const outputConfig = toAnthropicOutputConfig(responseFormat);
+    if (outputConfig) {
+      options.output_config = outputConfig;
+    }
     if (this.config.thinking?.type === "enabled") {
       options.thinking = {
         type: "enabled",
@@ -1485,12 +1608,14 @@ var OllamaAdapter = class {
       if (this.config.options) {
         Object.assign(ollamaOptions, this.config.options);
       }
+      const ollamaFormat = toOllamaFormat(request.config?.responseFormat);
       const payload = {
         model: request.config?.model || this.model,
         messages,
         tools,
         stream: true,
-        options: ollamaOptions
+        options: ollamaOptions,
+        ...ollamaFormat !== void 0 ? { format: ollamaFormat } : {}
       };
       logProviderPayload("ollama", "request payload", payload, request.debug);
       const response = await fetch(`${this.baseUrl}/api/chat`, {
@@ -1781,6 +1906,12 @@ var GoogleAdapter = class {
     }
     const messageId = generateMessageId();
     yield { type: "message:start", id: messageId };
+    const responseFormat = request.config?.responseFormat;
+    const geminiSchema = toGeminiSchema(responseFormat);
+    const responseFormatGenConfig = responseFormat ? {
+      responseMimeType: "application/json",
+      ...geminiSchema ? { responseJsonSchema: geminiSchema } : {}
+    } : {};
     try {
       logProviderPayload(
         "google",
@@ -1792,7 +1923,8 @@ var GoogleAdapter = class {
           tools: toolsArray.length > 0 ? toolsArray : void 0,
           generationConfig: {
             temperature: request.config?.temperature ?? this.config.temperature,
-            maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+            maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+            ...responseFormatGenConfig
           },
           messageParts: mergedContents[mergedContents.length - 1]?.parts
         },
@@ -1805,7 +1937,8 @@ var GoogleAdapter = class {
         tools: toolsArray.length > 0 ? toolsArray : void 0,
         generationConfig: {
           temperature: request.config?.temperature ?? this.config.temperature,
-          maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+          maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+          ...responseFormatGenConfig
         }
       });
       const lastMessage = mergedContents[mergedContents.length - 1];
@@ -1972,6 +2105,12 @@ var GoogleAdapter = class {
       }
     }
     const tools = formatToolsForGemini(request.actions);
+    const responseFormat = request.config?.responseFormat;
+    const geminiSchema = toGeminiSchema(responseFormat);
+    const responseFormatGenConfig = responseFormat ? {
+      responseMimeType: "application/json",
+      ...geminiSchema ? { responseJsonSchema: geminiSchema } : {}
+    } : {};
     const payload = {
       model: modelId,
       history: mergedContents.slice(0, -1),
@@ -1979,7 +2118,8 @@ var GoogleAdapter = class {
       tools: tools ? [tools] : void 0,
       generationConfig: {
         temperature: request.config?.temperature ?? this.config.temperature,
-        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...responseFormatGenConfig
       },
       messageParts: mergedContents[mergedContents.length - 1]?.parts
     };
@@ -1990,7 +2130,8 @@ var GoogleAdapter = class {
       tools: tools ? [tools] : void 0,
       generationConfig: {
         temperature: request.config?.temperature ?? this.config.temperature,
-        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens
+        maxOutputTokens: request.config?.maxTokens ?? this.config.maxTokens,
+        ...responseFormatGenConfig
       }
     });
     const lastMessage = mergedContents[mergedContents.length - 1];
@@ -2129,6 +2270,7 @@ var AzureAdapter = class {
         tools,
         temperature: request.config?.temperature ?? this.config.temperature,
         max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+        response_format: toOpenAIResponseFormat(request.config?.responseFormat),
         stream: true
       };
       logProviderPayload("azure", "request payload", payload, request.debug);
@@ -2228,7 +2370,8 @@ var AzureAdapter = class {
       messages,
       tools,
       temperature: request.config?.temperature ?? this.config.temperature,
-      max_tokens: request.config?.maxTokens ?? this.config.maxTokens
+      max_tokens: request.config?.maxTokens ?? this.config.maxTokens,
+      response_format: toOpenAIResponseFormat(request.config?.responseFormat)
     };
     logProviderPayload("azure", "request payload", payload, request.debug);
     const response = await client.chat.completions.create(payload);