plugin-ai-api 1.0.9 → 1.0.10

package/dist/externalVersion.js

@@ -10,14 +10,14 @@
 module.exports = {
   "react": "18.2.0",
   "antd": "5.24.2",
-  "@nocobase/client-v2": "2.1.23",
-  "@nocobase/flow-engine": "2.1.23",
-  "@nocobase/client": "2.1.23",
-  "@nocobase/plugin-acl": "2.1.23",
-  "@nocobase/server": "2.1.23",
+  "@nocobase/client-v2": "2.1.27",
+  "@nocobase/flow-engine": "2.1.27",
+  "@nocobase/client": "2.1.27",
+  "@nocobase/plugin-acl": "2.1.27",
+  "@nocobase/server": "2.1.27",
   "dayjs": "1.11.13",
-  "@nocobase/actions": "2.1.23",
-  "@nocobase/database": "2.1.23",
-  "@nocobase/resourcer": "2.1.23",
-  "@nocobase/plugin-ai": "2.1.23"
+  "@nocobase/actions": "2.1.27",
+  "@nocobase/database": "2.1.27",
+  "@nocobase/resourcer": "2.1.27",
+  "@nocobase/plugin-ai": "2.1.27"
 };

package/dist/server/routes/agent-completions.js

@@ -32,6 +32,7 @@ module.exports = __toCommonJS(agent_completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
 var import_role_permission = require("../middleware/role-permission");
+var import_streaming = require("../utils/streaming");
 var import_ai_employee_runtime = require("../utils/ai-employee-runtime");
 async function handleAgentCompletions(ctx, plugin) {
   var _a;
@@ -96,7 +97,7 @@ async function handleAgentCompletions(ctx, plugin) {
     );
     return;
   }
-  const wantStream = body.stream === true;
+  const wantStream = (0, import_streaming.isStreamingRequested)(body.stream);
   const lifecycle = (0, import_ai_employee_runtime.getAgentRuntimeLifecycle)(ctx);
   let runtimeContext;
   let lifecycleCompleted = false;
@@ -182,42 +183,87 @@ async function handleAgentCompletions(ctx, plugin) {
       );
       const originalWrite = ctx.res.write.bind(ctx.res);
       const originalEnd = ctx.res.end.bind(ctx.res);
+      const aiPlugin = ctx.app.pm.get("ai");
+      const abortAgent = () => {
+        var _a2, _b, _c;
+        if (!ctx.res.writableEnded) {
+          (_c = (_b = (_a2 = aiPlugin == null ? void 0 : aiPlugin.aiEmployeesManager) == null ? void 0 : _a2.conversationController) == null ? void 0 : _b.get(String(sessionId))) == null ? void 0 : _c.abort();
+        }
+      };
+      ctx.req.once("aborted", abortAgent);
+      ctx.res.once("close", abortAgent);
+      let streamSucceeded = false;
+      let sawToolCalls = false;
+      let pendingSse = "";
       ctx.res.end = (...args) => {
         if (args[0]) {
           ctx.res.write(args[0]);
         }
       };
       ctx.res.write = (data) => {
-        const text = typeof data === "string" ? data : data.toString("utf8");
-        for (const line of text.split("\n")) {
-          const trimmed = line.trim();
-          if (!trimmed.startsWith("data: ")) continue;
-          const jsonStr = trimmed.substring(6);
-          if (!jsonStr) continue;
-          try {
-            const event = JSON.parse(jsonStr);
-            if (event.type === "content" && event.body) {
-              originalWrite(
-                (0, import_openai_format.formatSSE)(
-                  (0, import_openai_format.toOpenAIStreamChunk)({
-                    id: completionId,
-                    model: body.model,
-                    delta: { content: String(event.body) }
+        var _a2;
+        pendingSse += typeof data === "string" ? data : data.toString("utf8");
+        const frames = pendingSse.split("\n\n");
+        pendingSse = frames.pop() || "";
+        for (const frame of frames) {
+          for (const line of frame.split("\n")) {
+            const trimmed = line.trim();
+            if (!trimmed.startsWith("data: ")) continue;
+            const jsonStr = trimmed.substring(6);
+            if (!jsonStr) continue;
+            try {
+              const event = JSON.parse(jsonStr);
+              if (event.type === "content" && event.body) {
+                originalWrite(
+                  (0, import_openai_format.formatSSE)(
+                    (0, import_openai_format.toOpenAIStreamChunk)({
+                      id: completionId,
+                      model: body.model,
+                      delta: { content: String(event.body) }
+                    })
+                  )
+                );
+              } else if (event.type === "tool_call_chunks" && Array.isArray(event.body)) {
+                const chunks = toOpenAIToolCallChunks(event.body);
+                if (chunks.length) {
+                  sawToolCalls = true;
+                  originalWrite(
+                    (0, import_openai_format.formatSSE)(
+                      (0, import_openai_format.toOpenAIStreamChunk)({
+                        id: completionId,
+                        model: body.model,
+                        delta: { tool_calls: chunks }
+                      })
+                    )
+                  );
+                }
+              } else if (!sawToolCalls && event.type === "tool_calls" && Array.isArray((_a2 = event.body) == null ? void 0 : _a2.toolCalls)) {
+                const chunks = toOpenAIToolCallChunks(event.body.toolCalls);
+                if (chunks.length) {
+                  sawToolCalls = true;
+                  originalWrite(
+                    (0, import_openai_format.formatSSE)(
+                      (0, import_openai_format.toOpenAIStreamChunk)({
+                        id: completionId,
+                        model: body.model,
+                        delta: { tool_calls: chunks }
+                      })
+                    )
+                  );
+                }
+              } else if (event.type === "error" && event.body) {
+                originalWrite(
+                  (0, import_openai_format.formatSSE)({
+                    error: {
+                      message: String(event.body),
+                      type: "server_error",
+                      code: "agent_error"
+                    }
                   })
-                )
-              );
-            } else if (event.type === "error" && event.body) {
-              originalWrite(
-                (0, import_openai_format.formatSSE)({
-                  error: {
-                    message: String(event.body),
-                    type: "server_error",
-                    code: "agent_error"
-                  }
-                })
-              );
+                );
+              }
+            } catch {
             }
-          } catch {
           }
         }
         return true;
@@ -229,7 +275,10 @@ async function handleAgentCompletions(ctx, plugin) {
             model: modelId
           })
         );
-        await aiEmployee.stream({ userMessages });
+        streamSucceeded = await aiEmployee.stream({ userMessages });
+        if (!streamSucceeded) {
+          throw new Error("AI Employee stream failed");
+        }
         try {
           await (lifecycle == null ? void 0 : lifecycle.runAfterHooks(runtimeContext, { succeeded: true }));
         } finally {
@@ -238,18 +287,25 @@ async function handleAgentCompletions(ctx, plugin) {
       } finally {
         ctx.res.write = originalWrite;
         ctx.res.end = originalEnd;
-        originalWrite(
-          (0, import_openai_format.formatSSE)(
-            (0, import_openai_format.toOpenAIStreamChunk)({
-              id: completionId,
-              model: body.model,
-              delta: {},
-              finishReason: "stop"
-            })
-          )
-        );
-        originalWrite((0, import_openai_format.formatSSEDone)());
-        originalEnd();
+        ctx.req.off("aborted", abortAgent);
+        ctx.res.off("close", abortAgent);
+        if (streamSucceeded && !ctx.res.destroyed) {
+          originalWrite(
+            (0, import_openai_format.formatSSE)(
+              (0, import_openai_format.toOpenAIStreamChunk)({
+                id: completionId,
+                model: body.model,
+                delta: {},
+                finishReason: "stop"
+              })
+            )
+          );
+          originalWrite((0, import_openai_format.formatSSEDone)());
+          ctx.state.aiApiStreamResult = { succeeded: true, id: completionId };
+        } else {
+          ctx.state.aiApiStreamResult = { succeeded: false, id: completionId, errorCode: "agent_error" };
+        }
+        if (!ctx.res.writableEnded && !ctx.res.destroyed) originalEnd();
       }
     } else {
       const aiEmployee = new AIEmployee(
@@ -312,6 +368,16 @@ async function handleAgentCompletions(ctx, plugin) {
     }
   }
 }
+function toOpenAIToolCallChunks(value) {
+  return value.map((call, fallbackIndex) => ({
+    index: typeof call.index === "number" ? call.index : fallbackIndex,
+    ...call.id ? { id: String(call.id), type: "function" } : {},
+    function: {
+      ...call.name ? { name: String(call.name) } : {},
+      ...call.args !== void 0 ? { arguments: typeof call.args === "string" ? call.args : JSON.stringify(call.args) } : {}
+    }
+  }));
+}
 function extractLastAiMessageContent(result) {
   var _a;
   if (!(result == null ? void 0 : result.messages) || !Array.isArray(result.messages)) return "";

package/dist/server/routes/chat-completions.js

@@ -31,6 +31,7 @@ __export(chat_completions_exports, {
 module.exports = __toCommonJS(chat_completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
+var import_streaming = require("../utils/streaming");
 var import_role_permission = require("../middleware/role-permission");
 async function handleChatCompletions(ctx, plugin) {
   var _a;
@@ -55,7 +56,7 @@ async function handleChatCompletions(ctx, plugin) {
     );
     return;
   }
-  const stream = body.stream === true;
+  const stream = (0, import_streaming.isStreamingRequested)(body.stream);
   const resolved = await (0, import_resolve_service.resolveModelString)(ctx, body.model);
   if (!resolved) {
     ctx.status = 404;
@@ -150,10 +151,22 @@ async function handleChatCompletions(ctx, plugin) {
     const langchainMessages = messages.map((msg) => {
       const role = msg.role === "assistant" ? "ai" : msg.role;
       const content = typeof msg.content === "string" ? msg.content : JSON.stringify(msg.content);
+      if (msg.role === "assistant" && msg.tool_calls) {
+        return {
+          role,
+          content,
+          tool_calls: msg.tool_calls,
+          additional_kwargs: { tool_calls: msg.tool_calls }
+        };
+      }
+      if (msg.role === "tool") {
+        return { role: "tool", content, tool_call_id: msg.tool_call_id, name: msg.name };
+      }
       return [role, content];
     });
     const completionId = (0, import_openai_format.generateCompletionId)();
-    const chatModel = provider.createModel();
+    const baseModel = provider.createModel();
+    const chatModel = bindRequestTools(baseModel, body.tools, body.tool_choice);
     if (stream) {
       await handleStreamingCompletion(ctx, chatModel, langchainMessages, completionId, body.model);
     } else {
@@ -163,7 +176,7 @@ async function handleChatCompletions(ctx, plugin) {
     ctx.log.error("AI API chat completions error:", err);
     if (!ctx.res.headersSent) {
       ctx.status = 500;
-      ctx.body = (0, import_openai_format.toOpenAIError)(500, err.message || "Internal server error", "server_error");
+      ctx.body = (0, import_openai_format.toOpenAIError)(500, getErrorMessage(err, "Internal server error"), "server_error");
     }
   }
 }
@@ -182,11 +195,13 @@ async function handleNonStreamingCompletion(ctx, chatModel, messages, completion
     total_tokens: result.usage_metadata.total_tokens || 0
   } : { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 };
   ctx.status = 200;
+  const toolCalls = normalizeToolCalls(result.tool_calls);
   ctx.body = (0, import_openai_format.toOpenAIResponse)({
     id: completionId,
     model: modelName,
     content,
-    usage
+    usage,
+    toolCalls
   });
 }
 async function handleStreamingCompletion(ctx, chatModel, messages, completionId, modelName) {
@@ -198,7 +213,8 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
     // Disable nginx buffering
   });
   ctx.status = 200;
-  ctx.res.write(
+  await (0, import_streaming.writeResponse)(
+    ctx,
     (0, import_openai_format.formatSSE)(
       (0, import_openai_format.toOpenAIStreamChunk)({
         id: completionId,
@@ -207,9 +223,13 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
       })
     )
   );
+  const requestAbort = (0, import_streaming.createRequestAbortController)(ctx);
+  let usage;
+  let finishReason = "stop";
   try {
-    const stream = await chatModel.stream(messages);
+    const stream = await chatModel.stream(messages, { signal: requestAbort.signal });
     for await (const chunk of stream) {
+      if (requestAbort.signal.aborted) throw requestAbort.signal.reason;
       let content = "";
       if (typeof chunk.content === "string") {
         content = chunk.content;
@@ -218,7 +238,8 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
         content = (textPart == null ? void 0 : textPart.text) || "";
       }
       if (content) {
-        ctx.res.write(
+        await (0, import_streaming.writeResponse)(
+          ctx,
           (0, import_openai_format.formatSSE)(
             (0, import_openai_format.toOpenAIStreamChunk)({
               id: completionId,
@@ -228,31 +249,94 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
           )
         );
       }
+      const toolCallChunks = normalizeToolCallChunks(chunk.tool_call_chunks);
+      if (toolCallChunks.length) {
+        finishReason = "tool_calls";
+        await (0, import_streaming.writeResponse)(
+          ctx,
+          (0, import_openai_format.formatSSE)((0, import_openai_format.toOpenAIStreamChunk)({ id: completionId, model: modelName, delta: { tool_calls: toolCallChunks } }))
+        );
+      }
+      if (chunk.usage_metadata) {
+        usage = {
+          prompt_tokens: chunk.usage_metadata.input_tokens || 0,
+          completion_tokens: chunk.usage_metadata.output_tokens || 0,
+          total_tokens: chunk.usage_metadata.total_tokens || 0
+        };
+      }
     }
-    ctx.res.write(
+    await (0, import_streaming.writeResponse)(
+      ctx,
       (0, import_openai_format.formatSSE)(
         (0, import_openai_format.toOpenAIStreamChunk)({
           id: completionId,
           model: modelName,
           delta: {},
-          finishReason: "stop"
+          finishReason
         })
       )
     );
-    ctx.res.write((0, import_openai_format.formatSSEDone)());
+    await (0, import_streaming.writeResponse)(ctx, (0, import_openai_format.formatSSEDone)());
+    ctx.state.aiApiStreamResult = { succeeded: true, id: completionId, usage };
   } catch (err) {
     ctx.log.error("AI API streaming error:", err);
-    ctx.res.write(
-      (0, import_openai_format.formatSSE)({
-        error: {
-          message: err.message || "Streaming error",
-          type: "server_error"
-        }
-      })
-    );
+    if (!ctx.res.destroyed && !ctx.res.writableEnded) {
+      await (0, import_streaming.writeResponse)(
+        ctx,
+        (0, import_openai_format.formatSSE)({
+          error: {
+            message: getErrorMessage(err, "Streaming error"),
+            type: "server_error"
+          }
+        })
+      );
+    }
+    ctx.state.aiApiStreamResult = { succeeded: false, id: completionId, usage, errorCode: "stream_error" };
   } finally {
-    ctx.res.end();
+    requestAbort.dispose();
+    if (!ctx.res.writableEnded && !ctx.res.destroyed) ctx.res.end();
+  }
+}
+function getErrorMessage(error, fallback) {
+  return error instanceof Error && error.message ? error.message : fallback;
+}
+function bindRequestTools(chatModel, tools, toolChoice) {
+  if (!Array.isArray(tools) || tools.length === 0) return chatModel;
+  if (typeof chatModel.bindTools !== "function") {
+    throw new Error("The selected LLM provider does not support tool calling");
   }
+  return chatModel.bindTools(tools, toolChoice === void 0 ? void 0 : { tool_choice: toolChoice });
+}
+function normalizeToolCalls(value) {
+  if (!Array.isArray(value) || value.length === 0) return void 0;
+  return value.map((call) => {
+    var _a, _b;
+    return {
+      id: String(call.id || ""),
+      type: "function",
+      function: {
+        name: String(call.name || ((_a = call.function) == null ? void 0 : _a.name) || ""),
+        arguments: serializeToolArguments(call.args ?? ((_b = call.function) == null ? void 0 : _b.arguments))
+      }
+    };
+  });
+}
+function normalizeToolCallChunks(value) {
+  if (!Array.isArray(value)) return [];
+  return value.map((call, fallbackIndex) => {
+    var _a, _b;
+    return {
+      index: typeof call.index === "number" ? call.index : fallbackIndex,
+      ...call.id ? { id: String(call.id), type: "function" } : {},
+      function: {
+        ...call.name ? { name: String(call.name) } : {},
+        ...call.args !== void 0 || ((_a = call.function) == null ? void 0 : _a.arguments) !== void 0 ? { arguments: serializeToolArguments(call.args ?? ((_b = call.function) == null ? void 0 : _b.arguments)) } : {}
+      }
+    };
+  });
+}
+function serializeToolArguments(value) {
+  return typeof value === "string" ? value : JSON.stringify(value ?? {});
 }
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {

package/dist/server/routes/completions.js

@@ -31,6 +31,7 @@ __export(completions_exports, {
 module.exports = __toCommonJS(completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
+var import_streaming = require("../utils/streaming");
 async function handleCompletions(ctx, plugin) {
   var _a;
   const body = ctx.request.body;
@@ -54,7 +55,7 @@ async function handleCompletions(ctx, plugin) {
     );
     return;
   }
-  const stream = body.stream === true;
+  const stream = (0, import_streaming.isStreamingRequested)(body.stream);
   const resolved = await (0, import_resolve_service.resolveModelString)(ctx, body.model);
   if (!resolved) {
     ctx.status = 404;
@@ -145,7 +146,7 @@ async function handleCompletions(ctx, plugin) {
     ctx.log.error("AI API completions error:", err);
     if (!ctx.res.headersSent) {
       ctx.status = 500;
-      ctx.body = (0, import_openai_format.toOpenAIError)(500, err.message || "Internal server error", "server_error");
+      ctx.body = (0, import_openai_format.toOpenAIError)(500, getErrorMessage(err, "Internal server error"), "server_error");
     }
   }
 }
@@ -189,9 +190,12 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
     "X-Accel-Buffering": "no"
   });
   ctx.status = 200;
+  const requestAbort = (0, import_streaming.createRequestAbortController)(ctx);
+  let usage;
   try {
-    const stream = await chatModel.stream(messages);
+    const stream = await chatModel.stream(messages, { signal: requestAbort.signal });
     for await (const chunk of stream) {
+      if (requestAbort.signal.aborted) throw requestAbort.signal.reason;
       let text = "";
       if (typeof chunk.content === "string") {
         text = chunk.content;
@@ -200,7 +204,8 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
         text = (textPart == null ? void 0 : textPart.text) || "";
       }
       if (text) {
-        ctx.res.write(
+        await (0, import_streaming.writeResponse)(
+          ctx,
           (0, import_openai_format.formatSSE)({
             id: completionId,
             object: "text_completion",
@@ -218,8 +223,16 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
           })
         );
       }
+      if (chunk.usage_metadata) {
+        usage = {
+          prompt_tokens: chunk.usage_metadata.input_tokens || 0,
+          completion_tokens: chunk.usage_metadata.output_tokens || 0,
+          total_tokens: chunk.usage_metadata.total_tokens || 0
+        };
+      }
     }
-    ctx.res.write(
+    await (0, import_streaming.writeResponse)(
+      ctx,
       (0, import_openai_format.formatSSE)({
         id: completionId,
         object: "text_completion",
@@ -236,21 +249,30 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
         ]
       })
     );
-    ctx.res.write((0, import_openai_format.formatSSEDone)());
+    await (0, import_streaming.writeResponse)(ctx, (0, import_openai_format.formatSSEDone)());
+    ctx.state.aiApiStreamResult = { succeeded: true, id: completionId, usage };
   } catch (err) {
     ctx.log.error("AI API completions streaming error:", err);
-    ctx.res.write(
-      (0, import_openai_format.formatSSE)({
-        error: {
-          message: err.message || "Streaming error",
-          type: "server_error"
-        }
-      })
-    );
+    if (!ctx.res.destroyed && !ctx.res.writableEnded) {
+      await (0, import_streaming.writeResponse)(
+        ctx,
+        (0, import_openai_format.formatSSE)({
+          error: {
+            message: getErrorMessage(err, "Streaming error"),
+            type: "server_error"
+          }
+        })
+      );
+    }
+    ctx.state.aiApiStreamResult = { succeeded: false, id: completionId, usage, errorCode: "stream_error" };
   } finally {
-    ctx.res.end();
+    requestAbort.dispose();
+    if (!ctx.res.writableEnded && !ctx.res.destroyed) ctx.res.end();
   }
 }
+function getErrorMessage(error, fallback) {
+  return error instanceof Error && error.message ? error.message : fallback;
+}
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   handleCompletions

package/dist/server/routes/router.js

@@ -54,7 +54,7 @@ const API_PREFIX = "/api/ai-llm/v1";
 function createAiLlmRouter(plugin) {
   const checkRateLimit = (0, import_rate_limit.createRateLimitMiddleware)(plugin.rateLimiter);
   return async (ctx, next) => {
-    var _a, _b;
+    var _a, _b, _c, _d;
     const { path, method } = ctx;
     if (!path.startsWith(API_PREFIX)) {
       return next();
@@ -117,7 +117,13 @@ function createAiLlmRouter(plugin) {
       if (method === "POST" && subPath === "/chat/completions") {
         const mode = await resolveMode(ctx);
         await (mode === "agent" ? (0, import_agent_completions.handleAgentCompletions)(ctx, plugin) : (0, import_chat_completions.handleChatCompletions)(ctx, plugin));
-        logRequest(ctx, requestId, model, "ok", Date.now() - t0);
+        logRequest(
+          ctx,
+          requestId,
+          model,
+          ((_c = ctx.state.aiApiStreamResult) == null ? void 0 : _c.succeeded) === false ? "error" : "ok",
+          Date.now() - t0
+        );
         return;
       }
       if (method === "POST" && subPath === "/embeddings") {
@@ -137,7 +143,13 @@ function createAiLlmRouter(plugin) {
         } else {
           await (0, import_completions.handleCompletions)(ctx, plugin);
         }
-        logRequest(ctx, requestId, model, "ok", Date.now() - t0);
+        logRequest(
+          ctx,
+          requestId,
+          model,
+          ((_d = ctx.state.aiApiStreamResult) == null ? void 0 : _d.succeeded) === false ? "error" : "ok",
+          Date.now() - t0
+        );
         return;
       }
       if (method === "GET" && subPath === "/models") {
@@ -177,7 +189,11 @@ function createAiLlmRouter(plugin) {
       logRequest(ctx, requestId, model, "error", Date.now() - t0);
       if (!ctx.res.headersSent) {
         ctx.status = 500;
-        ctx.body = (0, import_openai_format.toOpenAIError)(500, err.message || "Internal server error", "server_error");
+        ctx.body = (0, import_openai_format.toOpenAIError)(
+          500,
+          err instanceof Error && err.message ? err.message : "Internal server error",
+          "server_error"
+        );
       }
     } finally {
       if (usageId !== void 0) {

package/dist/server/usage.js

@@ -58,15 +58,16 @@ async function startUsageRecord(ctx, requestId, endpoint, model, streaming) {
 async function finishUsageRecord(ctx, id, startedAt, status) {
   var _a;
   const response = ctx.body || {};
-  const usage = response.usage;
+  const streamResult = ctx.state.aiApiStreamResult;
+  const usage = response.usage || (streamResult == null ? void 0 : streamResult.usage);
   const values = {
-    status,
+    status: streamResult ? streamResult.succeeded ? "succeeded" : "failed" : status,
     httpStatus: ctx.status,
-    errorCode: (_a = response.error) == null ? void 0 : _a.code,
+    errorCode: ((_a = response.error) == null ? void 0 : _a.code) || (streamResult == null ? void 0 : streamResult.errorCode),
     inputTokens: usage == null ? void 0 : usage.prompt_tokens,
     outputTokens: usage == null ? void 0 : usage.completion_tokens,
     totalTokens: usage == null ? void 0 : usage.total_tokens,
-    providerRequestId: response.id,
+    providerRequestId: response.id || (streamResult == null ? void 0 : streamResult.id),
     completedAt: /* @__PURE__ */ new Date(),
     durationMs: Date.now() - startedAt,
     responseMetadata: { usageSource: usage ? "response" : "unavailable" }

package/dist/server/utils/openai-format.js

@@ -71,7 +71,15 @@ function toOpenAIError(statusCode, message, type = "invalid_request_error", code
   };
 }
 function toOpenAIResponse(options) {
-  const { id, model, content, finishReason = "stop", usage } = options;
+  var _a;
+  const {
+    id,
+    model,
+    content,
+    finishReason = ((_a = options.toolCalls) == null ? void 0 : _a.length) ? "tool_calls" : "stop",
+    usage,
+    toolCalls
+  } = options;
   return {
     id,
     object: "chat.completion",
@@ -83,7 +91,8 @@ function toOpenAIResponse(options) {
         index: 0,
         message: {
           role: "assistant",
-          content
+          content,
+          ...(toolCalls == null ? void 0 : toolCalls.length) ? { tool_calls: toolCalls } : {}
         },
         logprobs: null,
         finish_reason: finishReason