diff --git a/ui/litellm-dashboard/src/components/chat/runChatRequest.test.ts b/ui/litellm-dashboard/src/components/chat/runChatRequest.test.ts
index 60ab8772112..aca9364c52b 100644
--- a/ui/litellm-dashboard/src/components/chat/runChatRequest.test.ts
+++ b/ui/litellm-dashboard/src/components/chat/runChatRequest.test.ts
@@ -1,5 +1,6 @@
 import OpenAI from "openai";
-import { beforeEach, describe, expect, it, vi } from "vitest";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import i18n from "@/i18n/i18n";
 import { makeOpenAIResponsesRequest } from "@/components/llm_calls/responses_api";
 import { generateChatImages } from "./imageHelpers";
 import { toResponsesMessages } from "./attachmentHelpers";
@@ -192,4 +193,67 @@ describe("runChatRequest", () => {
     expect(options.onUpdate).toHaveBeenCalledExactlyOnceWith({ content: " [stopped]" });
     expect(options.onResponseId).toHaveBeenCalledWith(null);
   });
+
+  describe("when the UI language is not English", () => {
+    beforeEach(async () => {
+      i18n.addResourceBundle(
+        "de",
+        "translation",
+        {
+          chat: {
+            runChatRequest: { genericError: "Fehler", fileInputHint: " Hinweis", researchNeedsBackground: "Hinweis" },
+          },
+        },
+        true,
+        true,
+      );
+      await i18n.changeLanguage("de");
+    });
+
+    afterEach(async () => {
+      await i18n.changeLanguage("en");
+      i18n.removeResourceBundle("de", "translation");
+    });
+
+    it("should keep the generic error in English because it is sent back to the model", async () => {
+      const options = requestOptions();
+      vi.mocked(makeOpenAIResponsesRequest).mockRejectedValue("offline");
+      await runChatRequest(options, i18n.t.bind(i18n));
+      expect(options.onUpdate).toHaveBeenLastCalledWith({ content: "Something went wrong. Please try again." });
+    });
+
+    it("should keep the file input hint in English because it is sent back to the model", async () => {
+      const options = requestOptions({
+        messages: [
+          {
+            role: "user",
+            content: "Read the report",
+            attachments: [
+              {
+                id: "report",
+                kind: "file",
+                name: "report.pdf",
+                mimeType: "application/pdf",
+                url: "data:application/pdf;base64,JVBERg==",
+              },
+            ],
+          },
+        ],
+      });
+      const error = new OpenAI.APIError(400, { message: "Provider detail" }, "Provider detail", new Headers());
+      vi.mocked(makeOpenAIResponsesRequest).mockRejectedValue(error);
+      await runChatRequest(options, i18n.t.bind(i18n));
+      expect(options.onUpdate).toHaveBeenLastCalledWith({
+        content: error.message + " Try a model that supports file inputs, or paste the relevant text.",
+      });
+    });
+
+    it("should keep the deep research notice in English because it is sent back to the model", async () => {
+      const options = requestOptions({ research: true });
+      await runChatRequest(options, i18n.t.bind(i18n));
+      expect(options.onUpdate).toHaveBeenCalledExactlyOnceWith({
+        content: "Enable background chats before starting deep research.",
+      });
+    });
+  });
 });
diff --git a/ui/litellm-dashboard/src/components/chat/runChatRequest.ts b/ui/litellm-dashboard/src/components/chat/runChatRequest.ts
index 7758f33254d..dc30f7d68f6 100644
--- a/ui/litellm-dashboard/src/components/chat/runChatRequest.ts
+++ b/ui/litellm-dashboard/src/components/chat/runChatRequest.ts
@@ -31,8 +31,8 @@ function requestErrorMessage(
   error: unknown,
   messages: ChatRequestOptions["messages"],
   mode: ChatRequestOptions["mode"],
-  t: TFunction = i18n.t.bind(i18n),
 ): string {
+  const t = i18n.getFixedT("en");
   const message = error instanceof Error ? error.message : t("chat.runChatRequest.genericError");
   const hasFile =
     mode === "chat" && messages.some((item) => item.attachments?.some((attachment) => attachment.kind === "file"));
@@ -67,7 +67,7 @@ export async function runChatRequest(
   try {
     signal.throwIfAborted();
     if (research && !background) {
-      onUpdate({ content: t("chat.runChatRequest.researchNeedsBackground") });
+      onUpdate({ content: i18n.getFixedT("en")("chat.runChatRequest.researchNeedsBackground") });
       return;
     }
     if (mode === "image") {
@@ -125,7 +125,7 @@ export async function runChatRequest(
   } catch (error: unknown) {
     onResponseId(null);
     const stopped = signal.aborted || (error instanceof Error && error.name === "AbortError");
-    const errorMessage = requestErrorMessage(error, messages, mode, t);
+    const errorMessage = requestErrorMessage(error, messages, mode);
     onUpdate({ content: stopped ? content + " [stopped]" : content || errorMessage });
   } finally {
     if (mcpEvents.length) onUpdate({ mcpEvents });
diff --git a/ui/litellm-dashboard/src/components/liteadmin/agent.test.ts b/ui/litellm-dashboard/src/components/liteadmin/agent.test.ts
index e9fae756427..80c21e81e64 100644
--- a/ui/litellm-dashboard/src/components/liteadmin/agent.test.ts
+++ b/ui/litellm-dashboard/src/components/liteadmin/agent.test.ts
@@ -2,6 +2,7 @@ import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import type { ChatCompletion, ChatCompletionCreateParamsNonStreaming } from "openai/resources/chat/completions";
 import { createGatewayClient } from "@/components/llm_calls/gateway_client";
 import { registerAuthHeaderNameGetter } from "@/lib/http/runtime";
+import i18n from "@/i18n/i18n";
 import {
   getPreferredLiteAdminModel,
   MAX_INPUT_LENGTH,
@@ -889,6 +890,42 @@ describe("SDK conversation and action lifecycle", () => {
   );
 });
 
+describe("LiteAdmin replies sent back to the model when the UI language is not English", () => {
+  beforeEach(async () => {
+    i18n.addResourceBundle(
+      "de",
+      "translation",
+      { aria: { agent: { stepLimit: "Schrittlimit", noAnswer: "Keine Antwort" } } },
+      true,
+      true,
+    );
+    await i18n.changeLanguage("de");
+  });
+
+  afterEach(async () => {
+    await i18n.changeLanguage("en");
+    i18n.removeResourceBundle("de", "translation");
+  });
+
+  it("keeps the step limit reply in English", async () => {
+    const input = options();
+    const model = transport(Array.from({ length: 6 }, () => completion([call("teams_list", teamsArgs)])));
+    await runLiteAdmin(input, model.client);
+    expect(input.onMessage).toHaveBeenCalledExactlyOnceWith(
+      "I reached the step limit. Any completed actions remain applied; check the relevant page before continuing.",
+    );
+  });
+
+  it("keeps the no answer reply in English", async () => {
+    const input = options();
+    const model = transport([completion([], null)]);
+    await runLiteAdmin(input, model.client);
+    expect(input.onMessage).toHaveBeenCalledExactlyOnceWith(
+      "The model returned no answer. Try another request or model.",
+    );
+  });
+});
+
 describe("inference destination", () => {
   const destinations = [
     ["", "/gateway", "https://dashboard.example/ui/", "https://dashboard.example/gateway", false],
diff --git a/ui/litellm-dashboard/src/components/liteadmin/agent.ts b/ui/litellm-dashboard/src/components/liteadmin/agent.ts
index b9b8420a6e5..9acce836323 100644
--- a/ui/litellm-dashboard/src/components/liteadmin/agent.ts
+++ b/ui/litellm-dashboard/src/components/liteadmin/agent.ts
@@ -141,8 +141,8 @@ export async function runLiteAdmin(options: LiteAdminOptions, client?: OpenAI):
   active();
   const message = completion.choices[0]?.message;
   if (message?.tool_calls?.length) {
-    options.onMessage(i18n.t("aria.agent.stepLimit"));
+    options.onMessage(i18n.getFixedT("en")("aria.agent.stepLimit"));
   } else if (!message?.content && !message?.refusal) {
-    options.onMessage(i18n.t("aria.agent.noAnswer"));
+    options.onMessage(i18n.getFixedT("en")("aria.agent.noAnswer"));
   }
 }