feat(live): forward safety_settings from generate_content_config to the Live API

src/google/adk/models/google_llm.py

@@ -453,6 +453,17 @@ async def connect(self, llm_request: LlmRequest) -> BaseLlmConnection:
             ' backend. Please use Vertex AI backend.'
         )
     llm_request.live_connect_config.tools = llm_request.config.tools
+    # Safety settings are configured via LlmAgent.generate_content_config, which
+    # only populates llm_request.config. Forward them so live runs honor the
+    # same safety configuration as non-live runs. An explicitly provided
+    # live_connect_config value takes precedence.
+    if (
+        llm_request.config.safety_settings is not None
+        and llm_request.live_connect_config.safety_settings is None
+    ):
+      llm_request.live_connect_config.safety_settings = (
+          llm_request.config.safety_settings
+      )
     logger.debug('Connecting to live with llm_request:%s', llm_request)
     logger.debug('Live connect config: %s', llm_request.live_connect_config)
     async with self._live_api_client.aio.live.connect(

tests/unittests/models/test_google_llm.py

@@ -852,6 +852,150 @@ async def __aexit__(self, *args):
         )
 
 
+@pytest.mark.asyncio
+async def test_connect_forwards_safety_settings(gemini_llm, llm_request):
+  """Live sessions receive safety_settings from generate_content_config."""
+  safety_settings = [
+      types.SafetySetting(
+          category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+          threshold=types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
+      ),
+      types.SafetySetting(
+          category=types.HarmCategory.HARM_CATEGORY_HARASSMENT,
+          threshold=types.HarmBlockThreshold.BLOCK_ONLY_HIGH,
+      ),
+  ]
+  llm_request.config.safety_settings = safety_settings
+  llm_request.live_connect_config = types.LiveConnectConfig()
+
+  mock_live_session = mock.AsyncMock()
+
+  with mock.patch.object(gemini_llm, "_live_api_client") as mock_live_client:
+
+    class MockLiveConnect:
+
+      async def __aenter__(self):
+        return mock_live_session
+
+      async def __aexit__(self, *args):
+        pass
+
+    mock_live_client.aio.live.connect.return_value = MockLiveConnect()
+
+    async with gemini_llm.connect(llm_request) as connection:
+      mock_live_client.aio.live.connect.assert_called_once()
+      config_arg = mock_live_client.aio.live.connect.call_args.kwargs["config"]
+
+      assert config_arg.safety_settings == safety_settings
+      assert isinstance(connection, GeminiLlmConnection)
+
+
+@pytest.mark.asyncio
+async def test_connect_keeps_existing_live_safety_settings(
+    gemini_llm, llm_request
+):
+  """An explicit live_connect_config.safety_settings is not overwritten."""
+  live_safety_settings = [
+      types.SafetySetting(
+          category=types.HarmCategory.HARM_CATEGORY_HATE_SPEECH,
+          threshold=types.HarmBlockThreshold.BLOCK_NONE,
+      ),
+  ]
+  llm_request.config.safety_settings = [
+      types.SafetySetting(
+          category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+          threshold=types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
+      ),
+  ]
+  llm_request.live_connect_config = types.LiveConnectConfig(
+      safety_settings=live_safety_settings
+  )
+
+  mock_live_session = mock.AsyncMock()
+
+  with mock.patch.object(gemini_llm, "_live_api_client") as mock_live_client:
+
+    class MockLiveConnect:
+
+      async def __aenter__(self):
+        return mock_live_session
+
+      async def __aexit__(self, *args):
+        pass
+
+    mock_live_client.aio.live.connect.return_value = MockLiveConnect()
+
+    async with gemini_llm.connect(llm_request):
+      config_arg = mock_live_client.aio.live.connect.call_args.kwargs["config"]
+
+      assert config_arg.safety_settings == live_safety_settings
+
+
+@pytest.mark.asyncio
+async def test_connect_keeps_empty_live_safety_settings(
+    gemini_llm, llm_request
+):
+  """An explicit empty live_connect_config.safety_settings is not overwritten.
+
+  An empty list means "send no safety settings" and is distinct from None,
+  which means "not configured here".
+  """
+  llm_request.config.safety_settings = [
+      types.SafetySetting(
+          category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+          threshold=types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
+      ),
+  ]
+  llm_request.live_connect_config = types.LiveConnectConfig(safety_settings=[])
+
+  mock_live_session = mock.AsyncMock()
+
+  with mock.patch.object(gemini_llm, "_live_api_client") as mock_live_client:
+
+    class MockLiveConnect:
+
+      async def __aenter__(self):
+        return mock_live_session
+
+      async def __aexit__(self, *args):
+        pass
+
+    mock_live_client.aio.live.connect.return_value = MockLiveConnect()
+
+    async with gemini_llm.connect(llm_request):
+      config_arg = mock_live_client.aio.live.connect.call_args.kwargs["config"]
+
+      assert config_arg.safety_settings is not None
+      assert len(config_arg.safety_settings) == 0
+
+
+@pytest.mark.asyncio
+async def test_connect_safety_settings_remain_none_when_unset(
+    gemini_llm, llm_request
+):
+  """No safety_settings anywhere leaves the live config untouched."""
+  llm_request.live_connect_config = types.LiveConnectConfig()
+
+  mock_live_session = mock.AsyncMock()
+
+  with mock.patch.object(gemini_llm, "_live_api_client") as mock_live_client:
+
+    class MockLiveConnect:
+
+      async def __aenter__(self):
+        return mock_live_session
+
+      async def __aexit__(self, *args):
+        pass
+
+    mock_live_client.aio.live.connect.return_value = MockLiveConnect()
+
+    async with gemini_llm.connect(llm_request):
+      config_arg = mock_live_client.aio.live.connect.call_args.kwargs["config"]
+
+      assert config_arg.safety_settings is None
+
+
 @pytest.mark.parametrize(
     (
         "api_backend, "