nuxt-app/server/api/llm/chat/index.post.ts

import { requireUser } from "#server/utils/context";
import { getProviderById, getModelById } from "#server/service/llm";
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { type LanguageModel, streamText, stepCountIs, convertToModelMessages } from 'ai';
import { getEnabledToolsForLlm } from "#server/service/agent-tool";
import log4js from "logger";

const logger = log4js.getLogger("APP");

function resolveModel(
  provider: {
    name: string;
    apiKey: string | null;
    baseUrl: string | null;
    parseMode: string;
  },
  modelId: string,
): LanguageModel {
  const baseUrl = provider.baseUrl?.replace(/\/+$/, "") || undefined;

  if (provider.parseMode === "anthropic") {
    throw createError({
      statusCode: 400,
      statusMessage: "Anthropic 解析模式暂不支持流式对话，请使用 OpenAI 兼容模式",
    });
  }

  const openaiCompatible = createOpenAICompatible({
    name: provider.name,
    apiKey: provider.apiKey || undefined,
    baseURL: baseUrl || "https://api.openai.com/v1",
  });

  return openaiCompatible(modelId) as LanguageModel;
}

export default defineEventHandler(async (event) => {
  const user = await requireUser(event);
  if (!user) {
    throw createError({ statusCode: 401, statusMessage: "未登录" });
  }

  const body = await readBody(event);
  const { modelId: llmModelId, messages, enableThinking, enableTools } = body as {
    modelId: number;
    messages: any[];
    enableThinking?: boolean;
    enableTools?: boolean;
  };

  if (!llmModelId || !messages || !Array.isArray(messages) || messages.length === 0) {
    throw createError({ statusCode: 400, statusMessage: "参数无效" });
  }

  const model = await getModelById(llmModelId, user.id);
  if (!model) {
    throw createError({ statusCode: 404, statusMessage: "模型不存在" });
  }

  const provider = await getProviderById(model.providerId, user.id);
  if (!provider) {
    throw createError({ statusCode: 404, statusMessage: "供应商不存在" });
  }

  if (provider.status !== "active") {
    throw createError({ statusCode: 400, statusMessage: "供应商已禁用" });
  }

  if (!provider.apiKey) {
    throw createError({ statusCode: 400, statusMessage: "供应商未配置 API Key" });
  }

  logger.info(
    "[%s] [LLM-CHAT] userId=%d modelId=%d provider=%s parseMode=%s messages=%d thinking=%s tools=%s",
    event.context.requestId ?? "-",
    user.id,
    llmModelId,
    provider.name,
    provider.parseMode,
    messages.length,
    enableThinking ? "on" : "off",
    enableTools ? "on" : "off",
  );

  const languageModel = resolveModel(provider, model.modelId);

  const { tools, approvalConfig } = enableTools ? await getEnabledToolsForLlm(user.id) : { tools: undefined, approvalConfig: {} };

  const modelMessages = await convertToModelMessages(messages);

  const result = streamText({
    model: languageModel,
    messages: modelMessages,
    maxOutputTokens: model.maxTokens || undefined,
    ...(tools && Object.keys(tools).length > 0
      ? { tools, stopWhen: stepCountIs(8), toolApproval: approvalConfig }
      : {}),
    ...(enableThinking
      ? {
          providerOptions: {
            openaiCompatible: { reasoningEffort: "high" },
          },
        }
      : {}),
    onError: (errorData) => {
      const errMsg = errorData?.error instanceof Error
        ? errorData.error.message
        : String(errorData?.error ?? "未知错误");
      logger.error("[%s] [LLM-CHAT] streamText error: %s", event.context.requestId ?? "-", errMsg);
    },
    onFinish: ({ finishReason, usage, steps }) => {
      logger.info(
        "[%s] [LLM-CHAT] finished: reason=%s steps=%d inputTokens=%d outputTokens=%d",
        event.context.requestId ?? "-",
        finishReason,
        steps.length,
        usage?.inputTokens ?? 0,
        usage?.outputTokens ?? 0,
      );
    },
  });

  return result.toUIMessageStreamResponse({ sendReasoning: true });
});