使用Vercel生成式UI调用Gemini API出现429配额超限问题求助

使用Vercel生成式UI搭配AI SDK，基于Next 14 + TypeScript开发，采用Google Gemini的gemini-1.5-pro-latest模型。本地运行完全正常，但部署到Vercel后返回429资源耗尽错误：

responseBody: '{
"error": {
"code": 429,
"message": "Resource has been exhausted (e.g. check quota).",
"status": "RESOURCE_EXHAUSTED"
}
}'

核心服务端调用函数代码如下：

export async function continueConversation(
  input: string
): Promise<ClientMessage> {
  "use server";

  const history = getMutableAIState();

  const result = await streamUI({
    model: google("models/gemini-1.5-pro-latest"),
    system: `
      You are a general purpose assistant, you can help the user with a variety of tasks. You can tell jokes, give place and song recommendations, and much more. You are a professional, don't use emote.
      `,
    messages: [...history.get(), { role: "user", content: input }],
    text: ({ content, done }) => {
      if (done) {
        history.done((messages: ServerMessage[]) => [
          ...messages,
          { role: "assistant", content },
        ]);
      }
      return (
        <article className="markdown-container">
          <Markdown remarkPlugins={[remarkGfm]}>{content}</Markdown>
        </article>
      );
    },
    tools: {
      getJoke: {
        description:
          "A tool when the user wants a joke. The joke should make the user laugh.",
        parameters: z.object({
          category: z.string().optional().describe("the category of the joke"),
        }),
        generate: async function* ({ category }) {
          yield <LoaderCircle />;
          const joke = await generateObject({
            model: google("models/gemini-1.5-pro-latest"),
            schema: jokeSchema,
            prompt:
              "Generate a joke that will make the user laugh. The joke should be in the category of " +
              category +
              ". If no category is provided, ask the user for a category.",
          });
          return <JokeComponent joke={joke.object} />;
        },
      },
      getPlaces: {
        description:
          "A tool when the user wants place recommendations based on the location and type.",
        parameters: z.object({
          location: z.string().describe("the user's location"),
          type: z.string().optional().describe("the type of place"),
        }),
        generate: async function* ({ location, type }) {
          yield <LoaderCircle className="loader-circle" />;
          const places = await generateObject({
            model: google("models/gemini-1.5-pro-latest"),
            schema: placeSchema,
            prompt:
              "Generate an array of places to visit in " +
              location +
              " with the type of " +
              (type || "any type") +
              ". The array should contain at least 5 places.",
          });
          if (places && places.object && Array.isArray(places.object)) {
            return <PlaceComponent place={places.object} />;
          } else {
            return <p>Something went wrong, please try again later.</p>;
          }
        },
      },
      getSongs: {
        description:
          "A tool when the user wants song recommendations based on the genre.",
        parameters: z.object({
          genre: z.string().optional().describe("the genre of the song"),
          singer: z.string().optional().describe("the singer of the song"),
        }),
        generate: async function* ({ genre, singer }) {
          yield <LoaderCircle />;
          const songs = await generateObject({
            model: google("models/gemini-1.5-pro-latest"),
            schema: songSchema,
            prompt:
              "Generate songs recommendation in the genre of " +
              (genre || "any genres") +
              "or by the singer " +
              (singer || "any singer") +
              ". Return an array of 3 songs.",
          });
          if (songs && songs.object && Array.isArray(songs.object)) {
            return <SongComponent song={songs.object} />;
          } else {
            return <p>Something went wrong, please try again later.</p>;
          }
        },
      },
    },
    
  });

  return {
    id: nanoid(),
    role: "assistant",
    display: result.value,
  };
}

已尝试更换多个新账号的API Key，问题依旧，但本地环境可正常运行，需要解决建议。

• 验证Vercel环境变量配置：确认Vercel项目中Google Gemini的API Key环境变量已正确设置，变量名与代码中读取的一致。可在服务端函数中添加日志验证变量是否存在（注意不要泄露密钥）。
• 检查Gemini模型配额：登录Google AI Studio查看gemini-1.5-pro-latest的配额使用情况，确认请求数、令牌数是否达上限。新账号的1.5 Pro模型配额较低，可临时切换到gemini-1.5-flash-latest模型测试，排查是否为配额问题。
• 优化API调用逻辑：当前工具函数（getJoke/getPlaces/getSongs）每次调用都会额外发起一次Gemini请求，单用户交互可能触发多次模型调用，加速配额消耗。可将工具功能整合到主prompt中，或添加请求节流机制，限制单位时间内的请求次数。
• 排查Vercel自身限流：Vercel的边缘函数/Serverless函数有请求频率限制，若部署后短时间内请求量较大，可能触发Vercel层面的限流。查看Vercel项目日志确认错误来源，必要时切换函数部署类型（如从Edge改为Serverless）或联系Vercel调整限额。
• 确认网络连通性：Vercel部署节点可能存在网络限制，导致无法正常访问Gemini API。可在服务端函数中添加网络诊断代码（如请求公开API），验证节点能否正常连接Google AI服务。

内容的提问来源于stack exchange，提问作者salm00n