fix(google): preserve request timeout with http options

livekit-plugins/livekit-plugins-google/livekit/plugins/google/llm.py

@@ -476,12 +476,16 @@ async def _run(self) -> None:
                     )
                 self._extra_kwargs.pop("tools", None)
                 self._extra_kwargs.pop("tool_config", None)
-            http_options = self._llm._opts.http_options or types.HttpOptions(
-                timeout=int(self._conn_options.timeout * 1000)
-            )
-            if not http_options.headers:
-                http_options.headers = {}
-            http_options.headers["x-goog-api-client"] = f"livekit-agents/{__version__}"
+            if is_given(self._llm._opts.http_options):
+                http_options = self._llm._opts.http_options.model_copy()
+                if http_options.timeout is None:
+                    http_options.timeout = int(self._conn_options.timeout * 1000)
+            else:
+                http_options = types.HttpOptions(timeout=int(self._conn_options.timeout * 1000))
+
+            headers = dict(http_options.headers or {})
+            headers["x-goog-api-client"] = f"livekit-agents/{__version__}"
+            http_options.headers = headers
             config = types.GenerateContentConfig(
                 system_instruction=(
                     None

tests/test_plugin_google_llm.py

@@ -7,6 +7,7 @@
 
 from livekit.agents import llm
 from livekit.agents.llm import ChatContext, function_tool
+from livekit.agents.types import APIConnectOptions
 from livekit.plugins.google.llm import LLM, LLMStream
 from livekit.plugins.google.realtime.realtime_api import RealtimeModel, RealtimeSession
 
@@ -243,6 +244,34 @@ async def example_tool(query: str) -> str:
         assert config.system_instruction is not None
         assert config.tools is not None and len(config.tools) >= 1
 
+    @pytest.mark.asyncio
+    async def test_request_merges_timeout_into_caller_http_options(self) -> None:
+        caller_http_options = types.HttpOptions(headers={"X-Vertex-Test": "1"})
+        llm = LLM(
+            model="gemini-2.5-flash",
+            api_key="test",
+            http_options=caller_http_options,
+        )
+
+        fake, captured = self._patched_stream_capture()
+        with patch.object(llm._client.aio.models, "generate_content_stream", fake):
+            stream = llm.chat(
+                chat_ctx=ChatContext.empty(),
+                conn_options=APIConnectOptions(timeout=7.5),
+            )
+            try:
+                async for _ in stream:
+                    pass
+            finally:
+                await stream.aclose()
+
+        config = captured["config"]
+        assert config.http_options.timeout == 7500
+        assert config.http_options.headers["X-Vertex-Test"] == "1"
+        assert "livekit-agents/" in config.http_options.headers["x-goog-api-client"]
+        assert caller_http_options.timeout is None
+        assert caller_http_options.headers == {"X-Vertex-Test": "1"}
+
 
 class TestMediaResolution:
     def test_llm_media_resolution_is_passed_to_stream_kwargs(self):