@yourgpt/llm-sdk 2.1.10-alpha.0 → 2.5.1-beta.0

package/dist/providers/togetherai/index.js

@@ -1,5 +1,234 @@
 'use strict';
 
+// src/adapters/base.ts
+function stringifyForDebug(value) {
+  return JSON.stringify(
+    value,
+    (_key, currentValue) => {
+      if (typeof currentValue === "bigint") {
+        return currentValue.toString();
+      }
+      if (currentValue instanceof Error) {
+        return {
+          name: currentValue.name,
+          message: currentValue.message,
+          stack: currentValue.stack
+        };
+      }
+      return currentValue;
+    },
+    2
+  );
+}
+function logProviderPayload(provider, label, payload, enabled) {
+  if (!enabled) {
+    return;
+  }
+  if (label.toLowerCase().includes("stream ")) {
+    return;
+  }
+  try {
+    console.log(
+      `[llm-sdk:${provider}] ${label}
+${stringifyForDebug(payload)}`
+    );
+  } catch (error) {
+    console.log(
+      `[llm-sdk:${provider}] ${label} (failed to stringify payload)`,
+      error
+    );
+  }
+}
+function parameterToJsonSchema(param) {
+  const schema = {
+    type: param.type
+  };
+  if (param.description) {
+    schema.description = param.description;
+  }
+  if (param.enum) {
+    schema.enum = param.enum;
+  }
+  if (param.type === "array" && param.items) {
+    schema.items = parameterToJsonSchema(
+      param.items
+    );
+  }
+  if (param.type === "object" && param.properties) {
+    schema.properties = Object.fromEntries(
+      Object.entries(param.properties).map(([key, prop]) => [
+        key,
+        parameterToJsonSchema(
+          prop
+        )
+      ])
+    );
+    schema.additionalProperties = false;
+  }
+  return schema;
+}
+function normalizeObjectJsonSchema(schema) {
+  if (!schema || typeof schema !== "object") {
+    return {
+      type: "object",
+      properties: {},
+      required: [],
+      additionalProperties: false
+    };
+  }
+  const normalized = { ...schema };
+  const type = normalized.type;
+  if (type === "object") {
+    const properties = normalized.properties && typeof normalized.properties === "object" && !Array.isArray(normalized.properties) ? normalized.properties : {};
+    normalized.properties = Object.fromEntries(
+      Object.entries(properties).map(([key, value]) => [
+        key,
+        normalizeObjectJsonSchema(value)
+      ])
+    );
+    const propertyKeys = Object.keys(properties);
+    const required = Array.isArray(normalized.required) ? normalized.required.filter(
+      (value) => typeof value === "string"
+    ) : [];
+    normalized.required = Array.from(/* @__PURE__ */ new Set([...required, ...propertyKeys]));
+    if (normalized.additionalProperties === void 0) {
+      normalized.additionalProperties = false;
+    }
+  } else if (type === "array" && normalized.items && typeof normalized.items === "object") {
+    normalized.items = normalizeObjectJsonSchema(
+      normalized.items
+    );
+  }
+  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 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
+  };
+}
+function formatTools(actions) {
+  return actions.map((action) => ({
+    type: "function",
+    function: {
+      name: action.name,
+      description: action.description,
+      parameters: {
+        type: "object",
+        properties: action.parameters ? Object.fromEntries(
+          Object.entries(action.parameters).map(([key, param]) => [
+            key,
+            parameterToJsonSchema(param)
+          ])
+        ) : {},
+        required: action.parameters ? Object.entries(action.parameters).filter(([, param]) => param.required).map(([key]) => key) : [],
+        additionalProperties: false
+      }
+    }
+  }));
+}
+function hasImageAttachments(message) {
+  const attachments = message.metadata?.attachments;
+  return attachments?.some((a) => a.type === "image") ?? false;
+}
+function attachmentToOpenAIImage(attachment) {
+  if (attachment.type !== "image") return null;
+  let imageUrl;
+  if (attachment.url) {
+    imageUrl = attachment.url;
+  } else if (attachment.data) {
+    imageUrl = attachment.data.startsWith("data:") ? attachment.data : `data:${attachment.mimeType || "image/png"};base64,${attachment.data}`;
+  } else {
+    return null;
+  }
+  return {
+    type: "image_url",
+    image_url: {
+      url: imageUrl,
+      detail: "auto"
+    }
+  };
+}
+function messageToOpenAIContent(message) {
+  const attachments = message.metadata?.attachments;
+  const content = message.content ?? "";
+  if (!hasImageAttachments(message)) {
+    return content;
+  }
+  const blocks = [];
+  if (content) {
+    blocks.push({ type: "text", text: content });
+  }
+  if (attachments) {
+    for (const attachment of attachments) {
+      const imageBlock = attachmentToOpenAIImage(attachment);
+      if (imageBlock) {
+        blocks.push(imageBlock);
+      }
+    }
+  }
+  return blocks;
+}
+function formatMessagesForOpenAI(messages, systemPrompt) {
+  const formatted = [];
+  if (systemPrompt) {
+    formatted.push({ role: "system", content: systemPrompt });
+  }
+  for (const msg of messages) {
+    if (msg.role === "system") {
+      formatted.push({ role: "system", content: msg.content ?? "" });
+    } else if (msg.role === "user") {
+      formatted.push({
+        role: "user",
+        content: messageToOpenAIContent(msg)
+      });
+    } else if (msg.role === "assistant") {
+      const hasToolCalls = msg.tool_calls && msg.tool_calls.length > 0;
+      const assistantMsg = {
+        role: "assistant",
+        // Gemini/xAI (OpenAI-compatible) reject content: "" on assistant messages with tool_calls
+        content: hasToolCalls ? msg.content || null : msg.content
+      };
+      if (hasToolCalls) {
+        assistantMsg.tool_calls = msg.tool_calls;
+      }
+      formatted.push(assistantMsg);
+    } else if (msg.role === "tool" && msg.tool_call_id) {
+      formatted.push({
+        role: "tool",
+        content: msg.content ?? "",
+        tool_call_id: msg.tool_call_id
+      });
+    }
+  }
+  return formatted;
+}
+
 // src/providers/togetherai/provider.ts
 function togetherai(modelId, options = {}) {
   const apiKey = options.apiKey ?? process.env.TOGETHER_API_KEY;
@@ -42,6 +271,10 @@ function togetherai(modelId, options = {}) {
       if (params.tools) {
         requestBody.tools = params.tools;
       }
+      const responseFormat = toOpenAIResponseFormat(params.responseFormat);
+      if (responseFormat) {
+        requestBody.response_format = responseFormat;
+      }
       const response = await client2.chat.completions.create(requestBody);
       const choice = response.choices[0];
       const message = choice.message;
@@ -77,6 +310,10 @@ function togetherai(modelId, options = {}) {
       if (params.tools) {
         requestBody.tools = params.tools;
       }
+      const responseFormat = toOpenAIResponseFormat(params.responseFormat);
+      if (responseFormat) {
+        requestBody.response_format = responseFormat;
+      }
       const stream = await client2.chat.completions.create(requestBody);
       const toolCallMap = /* @__PURE__ */ new Map();
       let totalPromptTokens = 0;
@@ -215,204 +452,6 @@ function generateToolCallId() {
   return generateId("call");
 }
 
-// src/adapters/base.ts
-function stringifyForDebug(value) {
-  return JSON.stringify(
-    value,
-    (_key, currentValue) => {
-      if (typeof currentValue === "bigint") {
-        return currentValue.toString();
-      }
-      if (currentValue instanceof Error) {
-        return {
-          name: currentValue.name,
-          message: currentValue.message,
-          stack: currentValue.stack
-        };
-      }
-      return currentValue;
-    },
-    2
-  );
-}
-function logProviderPayload(provider, label, payload, enabled) {
-  if (!enabled) {
-    return;
-  }
-  if (label.toLowerCase().includes("stream ")) {
-    return;
-  }
-  try {
-    console.log(
-      `[llm-sdk:${provider}] ${label}
-${stringifyForDebug(payload)}`
-    );
-  } catch (error) {
-    console.log(
-      `[llm-sdk:${provider}] ${label} (failed to stringify payload)`,
-      error
-    );
-  }
-}
-function parameterToJsonSchema(param) {
-  const schema = {
-    type: param.type
-  };
-  if (param.description) {
-    schema.description = param.description;
-  }
-  if (param.enum) {
-    schema.enum = param.enum;
-  }
-  if (param.type === "array" && param.items) {
-    schema.items = parameterToJsonSchema(
-      param.items
-    );
-  }
-  if (param.type === "object" && param.properties) {
-    schema.properties = Object.fromEntries(
-      Object.entries(param.properties).map(([key, prop]) => [
-        key,
-        parameterToJsonSchema(
-          prop
-        )
-      ])
-    );
-    schema.additionalProperties = false;
-  }
-  return schema;
-}
-function normalizeObjectJsonSchema(schema) {
-  if (!schema || typeof schema !== "object") {
-    return {
-      type: "object",
-      properties: {},
-      required: [],
-      additionalProperties: false
-    };
-  }
-  const normalized = { ...schema };
-  const type = normalized.type;
-  if (type === "object") {
-    const properties = normalized.properties && typeof normalized.properties === "object" && !Array.isArray(normalized.properties) ? normalized.properties : {};
-    normalized.properties = Object.fromEntries(
-      Object.entries(properties).map(([key, value]) => [
-        key,
-        normalizeObjectJsonSchema(value)
-      ])
-    );
-    const propertyKeys = Object.keys(properties);
-    const required = Array.isArray(normalized.required) ? normalized.required.filter(
-      (value) => typeof value === "string"
-    ) : [];
-    normalized.required = Array.from(/* @__PURE__ */ new Set([...required, ...propertyKeys]));
-    if (normalized.additionalProperties === void 0) {
-      normalized.additionalProperties = false;
-    }
-  } else if (type === "array" && normalized.items && typeof normalized.items === "object") {
-    normalized.items = normalizeObjectJsonSchema(
-      normalized.items
-    );
-  }
-  return normalized;
-}
-function formatTools(actions) {
-  return actions.map((action) => ({
-    type: "function",
-    function: {
-      name: action.name,
-      description: action.description,
-      parameters: {
-        type: "object",
-        properties: action.parameters ? Object.fromEntries(
-          Object.entries(action.parameters).map(([key, param]) => [
-            key,
-            parameterToJsonSchema(param)
-          ])
-        ) : {},
-        required: action.parameters ? Object.entries(action.parameters).filter(([, param]) => param.required).map(([key]) => key) : [],
-        additionalProperties: false
-      }
-    }
-  }));
-}
-function hasImageAttachments(message) {
-  const attachments = message.metadata?.attachments;
-  return attachments?.some((a) => a.type === "image") ?? false;
-}
-function attachmentToOpenAIImage(attachment) {
-  if (attachment.type !== "image") return null;
-  let imageUrl;
-  if (attachment.url) {
-    imageUrl = attachment.url;
-  } else if (attachment.data) {
-    imageUrl = attachment.data.startsWith("data:") ? attachment.data : `data:${attachment.mimeType || "image/png"};base64,${attachment.data}`;
-  } else {
-    return null;
-  }
-  return {
-    type: "image_url",
-    image_url: {
-      url: imageUrl,
-      detail: "auto"
-    }
-  };
-}
-function messageToOpenAIContent(message) {
-  const attachments = message.metadata?.attachments;
-  const content = message.content ?? "";
-  if (!hasImageAttachments(message)) {
-    return content;
-  }
-  const blocks = [];
-  if (content) {
-    blocks.push({ type: "text", text: content });
-  }
-  if (attachments) {
-    for (const attachment of attachments) {
-      const imageBlock = attachmentToOpenAIImage(attachment);
-      if (imageBlock) {
-        blocks.push(imageBlock);
-      }
-    }
-  }
-  return blocks;
-}
-function formatMessagesForOpenAI(messages, systemPrompt) {
-  const formatted = [];
-  if (systemPrompt) {
-    formatted.push({ role: "system", content: systemPrompt });
-  }
-  for (const msg of messages) {
-    if (msg.role === "system") {
-      formatted.push({ role: "system", content: msg.content ?? "" });
-    } else if (msg.role === "user") {
-      formatted.push({
-        role: "user",
-        content: messageToOpenAIContent(msg)
-      });
-    } else if (msg.role === "assistant") {
-      const hasToolCalls = msg.tool_calls && msg.tool_calls.length > 0;
-      const assistantMsg = {
-        role: "assistant",
-        // Gemini/xAI (OpenAI-compatible) reject content: "" on assistant messages with tool_calls
-        content: hasToolCalls ? msg.content || null : msg.content
-      };
-      if (hasToolCalls) {
-        assistantMsg.tool_calls = msg.tool_calls;
-      }
-      formatted.push(assistantMsg);
-    } else if (msg.role === "tool" && msg.tool_call_id) {
-      formatted.push({
-        role: "tool",
-        content: msg.content ?? "",
-        tool_call_id: msg.tool_call_id
-      });
-    }
-  }
-  return formatted;
-}
-
 // src/adapters/openai.ts
 var OpenAIAdapter = class _OpenAIAdapter {
   constructor(config) {
@@ -425,7 +464,6 @@ var OpenAIAdapter = class _OpenAIAdapter {
     if (baseUrl.includes("generativelanguage.googleapis.com")) return "google";
     if (baseUrl.includes("x.ai")) return "xai";
     if (baseUrl.includes("azure")) return "azure";
-    if (baseUrl.includes("openrouter.ai")) return "openrouter";
     return "openai";
   }
   async getClient() {
@@ -525,259 +563,12 @@ var OpenAIAdapter = class _OpenAIAdapter {
       rawResponse: response
     };
   }
-  /**
-   * OpenAI reasoning models on OpenRouter (o1/o3/o4/gpt-5 family) hide their
-   * reasoning content on the chat-completions endpoint. To surface reasoning
-   * SUMMARIES (not raw CoT, which OpenAI never exposes) we have to use the
-   * Responses API, which streams `response.reasoning_summary_text.delta` events.
-   *
-   * Match by prefix on the OpenRouter model id. Excludes openai/gpt-4o,
-   * openai/gpt-4.1, openai/chatgpt-* — those continue on chat-completions.
-   */
-  isOpenAIReasoningModelOnOpenRouter(activeModel) {
-    if (this.provider !== "openrouter") return false;
-    return activeModel.startsWith("openai/o1") || activeModel.startsWith("openai/o3") || activeModel.startsWith("openai/o4") || activeModel.startsWith("openai/gpt-5");
-  }
-  /**
-   * Convert ActionDefinition[] (the chat-completions tool shape used by the
-   * adapter) to the Responses API tool shape.
-   */
-  buildResponsesToolsFromActions(actions) {
-    if (!actions || actions.length === 0) return void 0;
-    const formatted = formatTools(actions);
-    return formatted.map((t) => ({
-      type: "function",
-      name: t.function.name,
-      description: t.function.description,
-      parameters: t.function.parameters
-    }));
-  }
-  /**
-   * Streaming Responses API path for OpenAI reasoning models on OpenRouter.
-   *
-   * Maps Responses API SSE events back to the same StreamEvent shapes the
-   * chat-completions path emits, so downstream consumers (processChunk.ts,
-   * frontend tool handlers, plan approval, specialist delegations) see
-   * identical events regardless of which path produced them.
-   *
-   *   response.reasoning_summary_text.delta  → thinking:start (once) + thinking:delta
-   *   response.output_text.delta             → message:delta
-   *   response.output_item.added (function_call) → action:start (queued buffer)
-   *   response.function_call_arguments.delta → action:args (progressive)
-   *   response.output_item.done (function_call) → final action:args + action:end
-   *   response.completed                     → message:end + done(usage)
-   *   response.error                         → error
-   */
-  async *streamWithResponsesAPI(request, activeModel, messageId) {
-    const client = await this.getClient();
-    const maxTokensValue = request.config?.maxTokens ?? this.config.maxTokens;
-    const payload = {
-      model: activeModel,
-      input: this.buildResponsesInput(request),
-      stream: true,
-      reasoning: {
-        effort: request.config?.reasoningEffort ?? "medium",
-        summary: "auto"
-      }
-    };
-    if (request.systemPrompt) payload.instructions = request.systemPrompt;
-    if (typeof maxTokensValue === "number")
-      payload.max_output_tokens = maxTokensValue;
-    const tools = this.buildResponsesToolsFromActions(request.actions);
-    if (tools && tools.length > 0) payload.tools = tools;
-    logProviderPayload(
-      "openai",
-      "responses-api request payload",
-      payload,
-      request.debug
-    );
-    let stream;
-    try {
-      stream = await client.responses.create(payload);
-    } catch (error) {
-      yield {
-        type: "error",
-        message: error instanceof Error ? error.message : "Unknown error",
-        code: "OPENAI_RESPONSES_ERROR"
-      };
-      return;
-    }
-    const toolBuffers = /* @__PURE__ */ new Map();
-    const itemIdToCallId = /* @__PURE__ */ new Map();
-    let usage;
-    let reasoningStarted = false;
-    let textStarted = false;
-    let finishEmitted = false;
-    const resolveCallId = (evt) => {
-      if (evt?.call_id) return evt.call_id;
-      if (evt?.item_id) return itemIdToCallId.get(evt.item_id) ?? evt.item_id;
-      if (evt?.item?.call_id) return evt.item.call_id;
-      if (evt?.item?.id) return evt.item.id;
-      return "";
-    };
-    try {
-      for await (const evt of stream) {
-        logProviderPayload(
-          "openai",
-          "responses-api stream chunk",
-          evt,
-          request.debug
-        );
-        if (request.signal?.aborted) break;
-        const t = evt?.type ?? "";
-        if (t === "response.reasoning_summary_text.delta") {
-          const delta = evt.delta ?? "";
-          if (!delta) continue;
-          if (!reasoningStarted) {
-            yield { type: "thinking:start" };
-            reasoningStarted = true;
-          }
-          yield { type: "thinking:delta", content: delta };
-          continue;
-        }
-        if (t === "response.reasoning_summary_text.done" || t === "response.reasoning.done") {
-          continue;
-        }
-        if (t === "response.output_text.delta") {
-          const text = evt.delta ?? "";
-          if (!text) continue;
-          if (reasoningStarted && !textStarted) {
-            yield { type: "thinking:end" };
-            textStarted = true;
-          }
-          yield { type: "message:delta", content: text };
-          continue;
-        }
-        if (t === "response.output_item.added") {
-          const item = evt.item;
-          if (item?.type === "function_call") {
-            const callId = item.call_id ?? item.id ?? "";
-            const itemId = item.id ?? callId;
-            if (callId) {
-              if (itemId && itemId !== callId) {
-                itemIdToCallId.set(itemId, callId);
-              }
-              if (!toolBuffers.has(callId)) {
-                toolBuffers.set(callId, {
-                  id: callId,
-                  name: item.name ?? "",
-                  arguments: item.arguments ?? "",
-                  emittedStart: false
-                });
-              }
-              const buf = toolBuffers.get(callId);
-              if (buf.name && !buf.emittedStart) {
-                yield { type: "action:start", id: buf.id, name: buf.name };
-                buf.emittedStart = true;
-              }
-            }
-          }
-          continue;
-        }
-        if (t === "response.function_call_arguments.delta") {
-          const callId = resolveCallId(evt);
-          const delta = evt.delta ?? "";
-          if (!callId || !delta) continue;
-          let buf = toolBuffers.get(callId);
-          if (!buf) {
-            buf = { id: callId, name: "", arguments: "", emittedStart: false };
-            toolBuffers.set(callId, buf);
-          }
-          buf.arguments += delta;
-          if (buf.emittedStart) {
-            yield {
-              type: "action:args",
-              id: buf.id,
-              args: buf.arguments
-            };
-          }
-          continue;
-        }
-        if (t === "response.output_item.done") {
-          const item = evt.item;
-          if (item?.type === "function_call") {
-            const callId = item.call_id ?? item.id ?? "";
-            const buf = toolBuffers.get(callId);
-            const name = buf?.name || item.name || "";
-            const argsStr = buf?.arguments || item.arguments || "{}";
-            if (callId && name) {
-              if (!buf?.emittedStart) {
-                yield { type: "action:start", id: callId, name };
-              }
-              yield {
-                type: "action:args",
-                id: callId,
-                args: argsStr
-              };
-              yield {
-                type: "action:end",
-                id: callId,
-                name
-              };
-            }
-            toolBuffers.delete(callId);
-          }
-          continue;
-        }
-        if (t === "response.completed") {
-          const u = evt.response?.usage;
-          if (u) {
-            usage = {
-              prompt_tokens: u.input_tokens ?? 0,
-              completion_tokens: u.output_tokens ?? 0,
-              total_tokens: u.total_tokens ?? (u.input_tokens ?? 0) + (u.output_tokens ?? 0)
-            };
-          }
-          for (const buf of toolBuffers.values()) {
-            if (!buf.id || !buf.name) continue;
-            if (!buf.emittedStart) {
-              yield { type: "action:start", id: buf.id, name: buf.name };
-            }
-            yield {
-              type: "action:args",
-              id: buf.id,
-              args: buf.arguments || "{}"
-            };
-            yield { type: "action:end", id: buf.id, name: buf.name };
-          }
-          toolBuffers.clear();
-          if (reasoningStarted && !textStarted) {
-            yield { type: "thinking:end" };
-          }
-          yield { type: "message:end" };
-          yield { type: "done", usage };
-          finishEmitted = true;
-          continue;
-        }
-        if (t === "response.error" || t === "error") {
-          const msg = evt.error?.message || evt.message || "Responses API error";
-          yield {
-            type: "error",
-            message: msg,
-            code: "OPENAI_RESPONSES_ERROR"
-          };
-          return;
-        }
-      }
-    } catch (error) {
-      yield {
-        type: "error",
-        message: error instanceof Error ? error.message : "Unknown error",
-        code: "OPENAI_RESPONSES_ERROR"
-      };
-      return;
-    }
-    if (!finishEmitted) {
-      if (reasoningStarted && !textStarted) {
-        yield { type: "thinking:end" };
-      }
-      yield { type: "message:end" };
-      yield { type: "done", usage };
-    }
-  }
   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,
@@ -787,6 +578,7 @@ 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);
@@ -908,37 +700,21 @@ var OpenAIAdapter = class _OpenAIAdapter {
           name: openaiToolOptions.toolChoice.name
         }
       } : openaiToolOptions?.toolChoice;
-      const isOpenRouter = this.provider === "openrouter";
-      const activeModel = request.config?.model || this.model;
-      const modelSlug = activeModel.replace("openai/", "");
-      const isOSeries = /^o[1-9]/.test(modelSlug);
-      const isOpenAIOnOpenRouter = isOpenRouter && activeModel.startsWith("openai/");
-      if (!this.config.disableThinking && this.isOpenAIReasoningModelOnOpenRouter(activeModel)) {
-        yield* this.streamWithResponsesAPI(request, activeModel, messageId);
-        return;
-      }
-      const maxTokensValue = request.config?.maxTokens ?? this.config.maxTokens;
+      const modelIdForPayload = request.config?.model || this.model;
       const payload = {
-        model: activeModel,
+        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,
+        ...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 },
-        // o-series: use max_completion_tokens + reasoning_effort, no temperature
-        // regular models: use max_tokens + temperature
-        ...isOSeries ? {
-          max_completion_tokens: maxTokensValue,
-          reasoning_effort: request.config?.reasoningEffort ?? "medium"
-        } : {
-          temperature: request.config?.temperature ?? this.config.temperature,
-          max_tokens: maxTokensValue
-        },
-        // Non-OpenAI OpenRouter models support OR's reasoning/include_reasoning params.
-        // When disableThinking=true we must explicitly send include_reasoning:false because
-        // models like Qwen3 and DeepSeek-R1 reason by default even without the reasoning param.
-        ...isOpenRouter && !isOpenAIOnOpenRouter ? this.config.disableThinking ? { include_reasoning: false } : { reasoning: { max_tokens: 8e3 }, include_reasoning: true } : {}
+        stream_options: { include_usage: true }
       };
       logProviderPayload("openai", "request payload", payload, request.debug);
       const stream = await client.chat.completions.create(payload);
@@ -946,7 +722,6 @@ var OpenAIAdapter = class _OpenAIAdapter {
       const collectedCitations = [];
       let citationIndex = 0;
       let usage;
-      let adapterReasoningStarted = false;
       for await (const chunk of stream) {
         logProviderPayload("openai", "stream chunk", chunk, request.debug);
         if (request.signal?.aborted) {
@@ -957,22 +732,6 @@ var OpenAIAdapter = class _OpenAIAdapter {
         if (delta?.content) {
           yield { type: "message:delta", content: delta.content };
         }
-        if (isOpenRouter) {
-          const rc = delta?.reasoning_content ?? delta?.reasoning ?? null;
-          if (rc) {
-            const rcText = typeof rc === "string" ? rc : Array.isArray(rc) && rc[0]?.text ? rc[0].text : "";
-            if (rcText) {
-              if (!adapterReasoningStarted) {
-                yield { type: "thinking:start" };
-                adapterReasoningStarted = true;
-              }
-              yield { type: "thinking:delta", content: rcText };
-            }
-          } else if (adapterReasoningStarted && (delta?.content || choice?.finish_reason)) {
-            yield { type: "thinking:end" };
-            adapterReasoningStarted = false;
-          }
-        }
         const annotations = delta?.annotations;
         if (annotations && annotations.length > 0) {
           for (const annotation of annotations) {
@@ -1020,11 +779,6 @@ var OpenAIAdapter = class _OpenAIAdapter {
               };
             } else if (currentToolCall && toolCall.function?.arguments) {
               currentToolCall.arguments += toolCall.function.arguments;
-              yield {
-                type: "action:args",
-                id: currentToolCall.id,
-                args: currentToolCall.arguments
-              };
             }
           }
         }
@@ -1100,24 +854,20 @@ var OpenAIAdapter = class _OpenAIAdapter {
         name: openaiToolOptions.toolChoice.name
       }
     } : openaiToolOptions?.toolChoice;
-    const activeModel2 = request.config?.model || this.model;
-    const modelSlug2 = activeModel2.replace("openai/", "");
-    const isOSeries2 = /^o[1-9]/.test(modelSlug2);
-    const maxTokensValue2 = request.config?.maxTokens ?? this.config.maxTokens;
+    const modelIdForCompletePayload = request.config?.model || this.model;
     const payload = {
-      model: activeModel2,
+      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,
-      stream: false,
-      ...isOSeries2 ? {
-        max_completion_tokens: maxTokensValue2,
-        reasoning_effort: request.config?.reasoningEffort ?? "medium"
-      } : {
-        temperature: request.config?.temperature ?? this.config.temperature,
-        max_tokens: maxTokensValue2
-      }
+      ...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);
     const response = await client.chat.completions.create(payload);