diff --git a/open-sse/translator/request/openai-to-gemini.ts b/open-sse/translator/request/openai-to-gemini.ts
index 39a7e16018..70e64fa197 100644
--- a/open-sse/translator/request/openai-to-gemini.ts
+++ b/open-sse/translator/request/openai-to-gemini.ts
@@ -238,6 +238,27 @@ function buildHistoricalToolResultContext(name: string, response: unknown): stri
   ].join("\n");
 }
 
+// Gemini-family APIs (incl. Antigravity / Vertex) reject a `contents[]` array that
+// has two adjacent entries with the same role:
+//   400 INVALID_ARGUMENT "Request contains consecutive messages with the same role".
+// Client history that carries consecutive user turns — or a tool-result turn (mapped
+// to role:"user") immediately followed by a plain user turn — would otherwise leak
+// that invalid alternation through. Merge adjacent same-role entries by concatenating
+// their parts, the same normalization the Kiro and Claude request paths already apply
+// (9router#2191).
+function mergeConsecutiveSameRoleContents(contents: GeminiContent[]): GeminiContent[] {
+  const merged: GeminiContent[] = [];
+  for (const entry of contents) {
+    const last = merged[merged.length - 1];
+    if (last && last.role === entry.role) {
+      last.parts.push(...entry.parts);
+    } else {
+      merged.push(entry);
+    }
+  }
+  return merged;
+}
+
 // Core: Convert OpenAI request to Gemini format (base for all variants)
 function openaiToGeminiBase(
   model: string,
@@ -585,6 +606,9 @@ function openaiToGeminiBase(
     }
   }
 
+  // Collapse any consecutive same-role contents Gemini would reject (9router#2191).
+  result.contents = mergeConsecutiveSameRoleContents(result.contents ?? []);
+
   // Convert tools
   const bodyTools = body.tools as Array<Record<string, unknown>> | undefined;
   const geminiTools = buildGeminiTools(bodyTools, {
diff --git a/tests/unit/translator-gemini-consecutive-role-2191.test.ts b/tests/unit/translator-gemini-consecutive-role-2191.test.ts
new file mode 100644
index 0000000000..66bc706b67
--- /dev/null
+++ b/tests/unit/translator-gemini-consecutive-role-2191.test.ts
@@ -0,0 +1,94 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+// Regression for 9router#2191: the OpenAI->Gemini request translator must not
+// emit two adjacent `contents[]` entries with the same role. Gemini-family APIs
+// (incl. Antigravity / Vertex) reject those with
+// 400 INVALID_ARGUMENT "Request contains consecutive messages with the same role".
+// The translator had no consecutive-same-role merge pass (unlike the Kiro and
+// Claude paths), so consecutive `user` turns — or a tool-result turn (role:user)
+// immediately followed by a plain user turn — produced an invalid alternation.
+
+const { openaiToGeminiRequest } = await import(
+  "../../open-sse/translator/request/openai-to-gemini.ts"
+);
+
+type GeminiContent = { role: string; parts: Array<Record<string, unknown>> };
+type GeminiReq = { contents: GeminiContent[] };
+
+function assertNoConsecutiveSameRole(contents: GeminiContent[], label: string) {
+  for (let i = 1; i < contents.length; i++) {
+    assert.notStrictEqual(
+      contents[i].role,
+      contents[i - 1].role,
+      `${label}: contents[${i - 1}] and contents[${i}] both have role "${contents[i].role}" ` +
+        `(Gemini rejects consecutive same-role messages)`
+    );
+  }
+}
+
+test("OpenAI -> Gemini merges two consecutive user messages into one content block", () => {
+  const body = {
+    messages: [
+      { role: "user", content: "Hello" },
+      { role: "user", content: "Additional context" },
+    ],
+  };
+  const result = openaiToGeminiRequest("gemini-2.5-pro", body, false) as GeminiReq;
+
+  assertNoConsecutiveSameRole(result.contents, "two-user");
+  // The two user turns collapse into a single user content carrying both parts.
+  assert.equal(result.contents.length, 1, "expected the two user turns to merge into one");
+  assert.equal(result.contents[0].role, "user");
+  const texts = result.contents[0].parts.map((p) => p.text);
+  assert.deepEqual(texts, ["Hello", "Additional context"]);
+});
+
+test("OpenAI -> Gemini does not emit a tool-result(user) turn adjacent to a user turn", () => {
+  // Agentic history: user -> assistant(tool_call) -> tool(result) -> user.
+  // The assistant block pushes model + user(toolResponse); the trailing plain
+  // user turn would otherwise produce two adjacent role:"user" contents.
+  const body = {
+    messages: [
+      { role: "user", content: "List files" },
+      {
+        role: "assistant",
+        content: null,
+        tool_calls: [
+          {
+            id: "call_1",
+            type: "function",
+            function: { name: "ls", arguments: '{"path":"."}' },
+          },
+        ],
+      },
+      { role: "tool", tool_call_id: "call_1", content: "a.ts\nb.ts" },
+      { role: "user", content: "Now read a.ts" },
+    ],
+  };
+  const result = openaiToGeminiRequest("gemini-2.5-pro", body, false) as GeminiReq;
+
+  assertNoConsecutiveSameRole(result.contents, "tool-result-then-user");
+  // Roles must strictly alternate: user, model, user (toolResp + "Now read a.ts" merged).
+  assert.deepEqual(
+    result.contents.map((c) => c.role),
+    ["user", "model", "user"]
+  );
+});
+
+test("OpenAI -> Gemini keeps a normally alternating conversation unchanged", () => {
+  const body = {
+    messages: [
+      { role: "user", content: "Hi" },
+      { role: "assistant", content: "Hello there" },
+      { role: "user", content: "How are you?" },
+    ],
+  };
+  const result = openaiToGeminiRequest("gemini-2.5-pro", body, false) as GeminiReq;
+
+  assertNoConsecutiveSameRole(result.contents, "alternating");
+  assert.deepEqual(
+    result.contents.map((c) => c.role),
+    ["user", "model", "user"]
+  );
+});