Cherry pick fixes to v1 (#5934)

Co-authored-by: asobran <asobran@google.com>

Author: wukath

src/google/adk/flows/llm_flows/base_llm_flow.py

@@ -48,6 +48,7 @@
 from ...tools.base_toolset import BaseToolset
 from ...tools.tool_context import ToolContext
 from ...utils.context_utils import Aclosing
+from ...utils import model_name_utils
 from .audio_cache_manager import AudioCacheManager
 from .functions import build_auth_request_event
 
@@ -516,6 +517,20 @@ async def run_live(
           )
           llm_request.live_connect_config.session_resumption.transparent = True
 
+        if (
+            isinstance(llm, Gemini)
+            and llm._api_backend == GoogleLLMVariant.GEMINI_API
+            and model_name_utils.is_gemini_3_1_flash_live(llm_request.model)
+            and llm_request.contents
+            and not invocation_context.live_session_resumption_handle
+        ):
+          if llm_request.live_connect_config is None:
+            llm_request.live_connect_config = types.LiveConnectConfig()
+          if llm_request.live_connect_config.history_config is None:
+            llm_request.live_connect_config.history_config = types.HistoryConfig(
+                initial_history_in_client_content=True
+            )
+
         logger.info(
             'Establishing live connection for agent: %s',
             invocation_context.agent.name,

src/google/adk/flows/llm_flows/basic.py

@@ -25,6 +25,7 @@
 from ...agents.invocation_context import InvocationContext
 from ...events.event import Event
 from ...models.llm_request import LlmRequest
+from ...utils import model_name_utils
 from ...utils.output_schema_utils import can_use_output_schema_with_tools
 from ._base_llm_processor import BaseLlmRequestProcessor
 
@@ -78,11 +79,13 @@ def _build_basic_request(
   llm_request.live_connect_config.realtime_input_config = (
       invocation_context.run_config.realtime_input_config
   )
+  active_model_name = getattr(getattr(agent, 'canonical_live_model', None), 'model', None) or llm_request.model
+  is_gemini_31 = model_name_utils.is_gemini_3_1_flash_live(active_model_name)
   llm_request.live_connect_config.enable_affective_dialog = (
-      invocation_context.run_config.enable_affective_dialog
+      None if is_gemini_31 else invocation_context.run_config.enable_affective_dialog
   )
   llm_request.live_connect_config.proactivity = (
-      invocation_context.run_config.proactivity
+      None if is_gemini_31 else invocation_context.run_config.proactivity
   )
   llm_request.live_connect_config.session_resumption = (
       invocation_context.run_config.session_resumption

src/google/adk/models/gemini_llm_connection.py

@@ -80,10 +80,24 @@ async def send_history(self, history: list[types.Content]):
     ]
 
     if contents:
+      is_gemini_31 = model_name_utils.is_gemini_3_1_flash_live(
+          self._model_version
+      )
+      # Gemini Enterprise Agent Platform does not support history_config in the SDK.
+      # To initialize a live session with prior history without hitting a 1007
+      # protocol error (invalid role mid-session), we consolidate previous multi-turn
+      # interactions into a unified contextual preamble on a single user role turn.
+      if is_gemini_31 and self._api_backend != GoogleLLMVariant.GEMINI_API:
+        collapsed_text = "Previous conversation history:\n"
+        for c in contents:
+          text_parts = "".join(p.text for p in c.parts if p.text)
+          collapsed_text += f'[{c.role}]: {text_parts}\n'
+        contents = [types.Content(role='user', parts=[types.Part.from_text(text=collapsed_text)])]
+
       logger.debug('Sending history to live connection: %s', contents)
       await self._gemini_session.send_client_content(
           turns=contents,
-          turn_complete=contents[-1].role == 'user',
+          turn_complete=True if is_gemini_31 else (contents[-1].role == 'user'),
       )
     else:
       logger.info('no content is sent')
@@ -159,14 +173,21 @@ async def send_realtime(self, input: RealtimeInput):
     else:
       raise ValueError('Unsupported input type: %s' % type(input))
 
-  def __build_full_text_response(self, text: str):
+  def __build_full_text_response(
+      self,
+      text: str,
+      is_thought: bool = False,
+      grounding_metadata: types.GroundingMetadata | None = None,
+  ):
     """Builds a full text response.
 
     The text should not be partial and the returned LlmResponse is not
     partial.
 
     Args:
       text: The text to be included in the response.
+      is_thought: Whether the text is a thought.
+      grounding_metadata: The grounding metadata to include.
 
     Returns:
       An LlmResponse containing the full text.
@@ -176,6 +197,8 @@ def __build_full_text_response(self, text: str):
             role='model',
             parts=[types.Part.from_text(text=text)],
         ),
+        grounding_metadata=grounding_metadata,
+        partial=False,
         live_session_id=self._gemini_session.session_id,
     )
 
@@ -188,6 +211,7 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
 
     text = ''
     tool_call_parts = []
+    pending_grounding_metadata = None
     async with Aclosing(self._gemini_session.receive()) as agen:
       # TODO(b/440101573): Reuse StreamingResponseAggregator to accumulate
       # partial content and emit responses as needed.
@@ -203,6 +227,10 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
           )
         if message.server_content:
           content = message.server_content.model_turn
+          if message.server_content.grounding_metadata:
+            pending_grounding_metadata = (
+                message.server_content.grounding_metadata
+            )
 
           # Standalone grounding_metadata event (when content is empty)
           if (
@@ -215,6 +243,9 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
                 interrupted=message.server_content.interrupted,
                 model_version=self._model_version,
                 live_session_id=live_session_id,
+                turn_complete_reason=getattr(
+                    message.server_content, 'turn_complete_reason', None
+                ),
             )
 
           if content and content.parts:
@@ -223,19 +254,31 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
                 interrupted=message.server_content.interrupted,
                 model_version=self._model_version,
                 live_session_id=live_session_id,
+                turn_complete_reason=getattr(
+                    message.server_content, 'turn_complete_reason', None
+                ),
             )
             # grounding_metadata is yielded again at turn_complete,
             # so avoid duplicating it here if turn_complete is true.
             if not message.server_content.turn_complete:
               llm_response.grounding_metadata = (
                   message.server_content.grounding_metadata
               )
-            if content.parts[0].text:
-              text += content.parts[0].text
-              llm_response.partial = True
+            has_inline_data = any(p.inline_data for p in content.parts)
+            for part in content.parts:
+              if part.text:
+                current_is_thought = getattr(part, 'thought', False)
+                if text and current_is_thought != is_thought:
+                  yield self.__build_full_text_response(text, is_thought)
+                  text = ''
+                  is_thought = False
+
+                text += part.text
+                is_thought = current_is_thought
+                llm_response.partial = True
             # don't yield the merged text event when receiving audio data
-            elif text and not content.parts[0].inline_data:
-              yield self.__build_full_text_response(text)
+            if text and not any(p.text for p in content.parts) and not has_inline_data:
+              yield self.__build_full_text_response(text, is_thought)
               text = ''
             yield llm_response
           # Note: in some cases, tool_call may arrive before
@@ -324,9 +367,14 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
               )
               self._output_transcription_text = ''
           if message.server_content.turn_complete:
+            g_metadata_to_yield = pending_grounding_metadata
             if text:
-              yield self.__build_full_text_response(text)
+              yield self.__build_full_text_response(
+                  text, is_thought, g_metadata_to_yield
+              )
               text = ''
+              is_thought = False
+              g_metadata_to_yield = None
             if tool_call_parts:
               logger.debug('Returning aggregated tool_call_parts')
               yield LlmResponse(
@@ -338,9 +386,13 @@ async def receive(self) -> AsyncGenerator[LlmResponse, None]:
             yield LlmResponse(
                 turn_complete=True,
                 interrupted=message.server_content.interrupted,
-                grounding_metadata=message.server_content.grounding_metadata,
+                grounding_metadata=message.server_content.grounding_metadata
+                or g_metadata_to_yield,
                 model_version=self._model_version,
                 live_session_id=live_session_id,
+                turn_complete_reason=getattr(
+                    message.server_content, 'turn_complete_reason', None
+                ),
             )
             break
           # in case of empty content or parts, we still surface it

src/google/adk/models/llm_response.py

@@ -81,6 +81,12 @@ class LlmResponse(BaseModel):
   Only used for streaming mode.
   """
 
+  turn_complete_reason: Optional[types.TurnCompleteReason] = None
+  """The reason why the turn is complete.
+
+  Only used for streaming mode.
+  """
+
   finish_reason: Optional[types.FinishReason] = None
   """The finish reason of the response."""