diff --git a/app/[locale]/docs/[...slug]/page.tsx b/app/[locale]/docs/[...slug]/page.tsx
index f8c79764..99668ff8 100644
--- a/app/[locale]/docs/[...slug]/page.tsx
+++ b/app/[locale]/docs/[...slug]/page.tsx
@@ -214,6 +214,7 @@ export default async function DocPage({ params }: Param) {
           title: page.data.title,
           description: page.data.description,
           slug: slug?.join("/"),
+          locale,
         }}
       />
     </>
diff --git a/app/api/chat/route.ts b/app/api/chat/route.ts
index fca7d3e3..97dba3a2 100644
--- a/app/api/chat/route.ts
+++ b/app/api/chat/route.ts
@@ -2,6 +2,8 @@ import { streamText, UIMessage, convertToModelMessages } from "ai";
 import { getModel, requiresApiKey, type AIProvider } from "@/lib/ai/models";
 import { buildSystemMessage } from "@/lib/ai/prompt";
 import { source } from "@/lib/source";
+import { routing } from "@/i18n/routing";
+import { hasLocale } from "next-intl";
 import { limitChat, rateLimitResponse } from "@/lib/rate-limit";
 import fs from "fs/promises";
 import path from "path";
@@ -20,6 +22,7 @@ interface ChatRequest {
     description?: string;
     content?: string;
     slug?: string;
+    locale?: string;
   };
   provider?: AIProvider;
   apiKey?: string;
@@ -140,7 +143,11 @@ export async function POST(req: Request) {
     if (pageContext?.slug && !pageContext.content) {
       try {
         const slugArray = pageContext.slug.split("/");
-        const page = source.getPage(slugArray);
+        // 不传 locale 会固定拿到 zh 原文，/en 页的助手就读中文；非法值交给 fallback（zh）
+        const locale = hasLocale(routing.locales, pageContext.locale)
+          ? pageContext.locale
+          : undefined;
+        const page = source.getPage(slugArray, locale);
 
         if (page) {
           const cachedContent = mdxContentCache.get(page.path);
diff --git a/app/api/suggestions/route.ts b/app/api/suggestions/route.ts
index 2fa23003..5c045d79 100644
--- a/app/api/suggestions/route.ts
+++ b/app/api/suggestions/route.ts
@@ -120,7 +120,7 @@ export async function POST(req: Request) {
 
     let text = "";
     // 判断是否可以使用缓存（如果使用了自定义 apiKey，为了防止数据交叉，不采用公用缓存）
-    if (provider !== "intern" && !process.env.ZHIPU_API_KEY) {
+    if (provider !== "intern") {
       const { text: directText } = await generateText({
         model,
         prompt,
@@ -128,12 +128,7 @@ export async function POST(req: Request) {
       text = directText;
     } else {
       // 确定内部用于缓存匹配的模型标识
-      const internalModelId =
-        provider === "intern"
-          ? process.env.ZHIPU_API_KEY
-            ? "glm"
-            : "intern"
-          : "intern";
+      const internalModelId = process.env.ZHIPU_API_KEY ? "glm" : "intern";
       // 将缓存粒度关联到请求本身的内容上
       const suggestionKey = isWelcomeRequest
         ? `welcome-${pageContext?.slug || "default"}`
diff --git a/app/components/DocsAssistant.tsx b/app/components/DocsAssistant.tsx
index d6f4e4ab..18c4a012 100644
--- a/app/components/DocsAssistant.tsx
+++ b/app/components/DocsAssistant.tsx
@@ -19,6 +19,7 @@ interface PageContext {
   title?: string;
   description?: string;
   slug?: string;
+  locale?: string;
 }
 
 export interface WelcomeSuggestion {
diff --git a/tests/suggestions-route.test.ts b/tests/suggestions-route.test.ts
new file mode 100644
index 00000000..f2fc3b5c
--- /dev/null
+++ b/tests/suggestions-route.test.ts
@@ -0,0 +1,67 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+const generateText = vi.fn(async () => ({ text: '["q1","q2","q3"]' }));
+const unstableCache = vi.fn(
+  (fn: (...args: unknown[]) => unknown, _keys?: string[], _opts?: object) =>
+    (...args: unknown[]) =>
+      fn(...args),
+);
+
+vi.mock("ai", () => ({ generateText }));
+vi.mock("next/cache", () => ({ unstable_cache: unstableCache }));
+vi.mock("@/lib/rate-limit", () => ({
+  limitChat: async () => ({ success: true }),
+  rateLimitResponse: () => new Response(null, { status: 429 }),
+}));
+vi.mock("@/lib/ai/models", () => ({
+  getModel: (provider: string, apiKey?: string) => ({ provider, apiKey }),
+  requiresApiKey: (provider: string) => provider !== "intern",
+}));
+vi.mock("@/lib/ai/providers/glm", () => ({
+  createGlmFlashModel: () => ({ provider: "glm" }),
+}));
+
+const { POST } = await import("@/app/api/suggestions/route");
+
+// 公共缓存按请求建一层 unstable_cache，key 形如 suggestion-key-<...>
+const usedSharedCache = () =>
+  unstableCache.mock.calls.some(([, keys]) =>
+    (keys as string[]).some((k) => k.startsWith("suggestion-key-")),
+  );
+
+function post(body: object) {
+  return POST(
+    new Request("http://localhost/api/suggestions", {
+      method: "POST",
+      body: JSON.stringify(body),
+    }),
+  );
+}
+
+describe("/api/suggestions 模型选择", () => {
+  beforeEach(() => vi.stubEnv("ZHIPU_API_KEY", "server-key"));
+  afterEach(() => {
+    vi.unstubAllEnvs();
+    vi.clearAllMocks();
+  });
+
+  it("自带 key 的 provider 用用户自己的模型，不进公共缓存（即使服务端配了 ZHIPU）", async () => {
+    await post({ messages: [], provider: "openai", apiKey: "user-key" });
+
+    expect(usedSharedCache()).toBe(false);
+    expect(generateText).toHaveBeenCalledWith(
+      expect.objectContaining({
+        model: { provider: "openai", apiKey: "user-key" },
+      }),
+    );
+  });
+
+  it("默认 intern 走公共缓存，配了 ZHIPU 时用 GLM", async () => {
+    await post({ messages: [] });
+
+    expect(usedSharedCache()).toBe(true);
+    expect(generateText).toHaveBeenCalledWith(
+      expect.objectContaining({ model: { provider: "glm" } }),
+    );
+  });
+});