[pull] main from google:main

contributing/samples/api_registry_agent/agent.py

@@ -26,7 +26,7 @@
     mcp_server_name=MCP_SERVER_NAME,
 )
 root_agent = LlmAgent(
-    model="gemini-2.0-flash",
+    model="gemini-2.5-flash",
     name="bigquery_assistant",
     instruction=f"""
 You are a helpful data analyst assistant with access to BigQuery. The project ID is: {PROJECT_ID}

contributing/samples/application_integration_agent/README.md

@@ -7,7 +7,7 @@ This sample demonstrates how to use the `ApplicationIntegrationToolset` within a
 ## Prerequisites
 
 1.  **Set up Integration Connection:**
-    *   You need an existing [Integration connection](https://cloud.google.com/integration-connectors/docs/overview) configured to interact with your Jira instance. Follow the [documentation](https://google.github.io/adk-docs/tools/google-cloud-tools/#use-integration-connectors) to provision the Integration Connector in Google Cloud and then use this [documentation](https://cloud.google.com/integration-connectors/docs/connectors/jiracloud/configure) to create an Jira connection. Note the `Connection Name`, `Project ID`, and `Location` of your connection.
+    *   You need an existing [Integration connection](https://cloud.google.com/integration-connectors/docs/overview) configured to interact with your Jira instance. Follow the [documentation](https://google.github.io/adk-docs/tools/google-cloud-tools/#use-integration-connectors) to provision the Integration Connector in Google Cloud and then use this [documentation](https://cloud.google.com/integration-connectors/docs/connectors/jiracloud/configure) to create a Jira connection. Note the `Connection Name`, `Project ID`, and `Location` of your connection.
     * 
 
 2.  **Configure Environment Variables:**

contributing/samples/mcp_dynamic_header_agent/agent.py

@@ -18,7 +18,7 @@
 from google.adk.tools.mcp_tool.mcp_toolset import McpToolset
 
 root_agent = LlmAgent(
-    model='gemini-2.0-flash',
+    model='gemini-2.5-flash',
     name='tenant_agent',
     instruction="""You are a helpful assistant that helps users get tenant
  information. Call the get_tenant_data tool when the user asks for tenant data.""",

contributing/samples/mcp_postgres_agent/agent.py

@@ -31,7 +31,7 @@
   )
 
 root_agent = LlmAgent(
-    model="gemini-2.0-flash",
+    model="gemini-2.5-flash",
     name="postgres_agent",
     instruction=(
         "You are a PostgreSQL database assistant. "

contributing/samples/mcp_service_account_agent/agent.py

@@ -29,7 +29,7 @@
 SCOPES = {"https://www.googleapis.com/auth/cloud-platform": ""}
 
 root_agent = LlmAgent(
-    model="gemini-2.0-flash",
+    model="gemini-2.5-flash",
     name="enterprise_assistant",
     instruction="""
 Help the user with the tools available to you.

contributing/samples/mcp_sse_agent/agent.py

@@ -28,7 +28,7 @@
 )
 
 root_agent = LlmAgent(
-    model='gemini-2.0-flash',
+    model='gemini-2.5-flash',
     name='enterprise_assistant',
     instruction=McpInstructionProvider(
         connection_params=connection_params,

contributing/samples/mcp_stdio_notion_agent/agent.py

@@ -29,7 +29,7 @@
 })
 
 root_agent = LlmAgent(
-    model="gemini-2.0-flash",
+    model="gemini-2.5-flash",
     name="notion_agent",
     instruction=(
         "You are my workspace assistant. "

contributing/samples/mcp_stdio_server_agent/agent.py

@@ -23,7 +23,7 @@
 _allowed_path = os.path.dirname(os.path.abspath(__file__))
 
 root_agent = LlmAgent(
-    model='gemini-2.0-flash',
+    model='gemini-2.5-flash',
     name='enterprise_assistant',
     instruction=f"""\
 Help user accessing their file systems.

contributing/samples/mcp_streamablehttp_agent/agent.py

@@ -22,7 +22,7 @@
 _allowed_path = os.path.dirname(os.path.abspath(__file__))
 
 root_agent = LlmAgent(
-    model='gemini-2.0-flash',
+    model='gemini-2.5-flash',
     name='enterprise_assistant',
     instruction=f"""\
 Help user accessing their file systems.

contributing/samples/multi_agent_seq_config/README.md

@@ -6,7 +6,7 @@ The whole process is:
 
 1. An agent backed by a cheap and fast model to write initial version.
 2. An agent backed by a smarter and a little more expensive to review the code.
-3. An final agent backed by the smartest and slowest model to write the final revision.
+3. A final agent backed by the smartest and slowest model to write the final revision.
 
 Sample queries:
 

contributing/samples/spanner_rag_agent/README.md

@@ -181,7 +181,7 @@ type.
 
 ## 💬 Sample prompts
 
-* I'd like to buy a starter bike for my 3 year old child, can you show me the recommendation?
+* I'd like to buy a starter bike for my 3-year-old child, can you show me the recommendation?
 
 ![Spanner RAG Sample Agent](Spanner_RAG_Sample_Agent.png)
 

pyproject.toml

@@ -123,7 +123,7 @@ test = [
   "kubernetes>=29.0.0",                           # For GkeCodeExecutor
   "langchain-community>=0.3.17",
   "langgraph>=0.2.60, <0.4.8",                    # For LangGraphAgent
-  "litellm>=1.75.5, <2.0.0",                      # For LiteLLM tests
+  "litellm>=1.75.5, <1.81.0",                     # For LiteLLM tests
   "llama-index-readers-file>=0.4.0",              # For retrieval tests
   "openai>=1.100.2",                              # For LiteLLM
   "pytest-asyncio>=0.25.0",
@@ -153,7 +153,7 @@ extensions = [
   "docker>=7.0.0",                              # For ContainerCodeExecutor
   "kubernetes>=29.0.0",                         # For GkeCodeExecutor
   "langgraph>=0.2.60, <0.4.8",                  # For LangGraphAgent
-  "litellm>=1.75.5",                            # For LiteLlm class. Currently has OpenAI limitations. TODO: once LiteLlm fix it
+  "litellm>=1.75.5, <1.81.0",                   # For LiteLlm class. Currently has OpenAI limitations. TODO: once LiteLlm fix it
   "llama-index-readers-file>=0.4.0",            # For retrieval using LlamaIndex.
   "llama-index-embeddings-google-genai>=0.3.0", # For files retrieval using LlamaIndex.
   "lxml>=5.3.0",                                # For load_web_page tool.

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

@@ -38,8 +38,6 @@
 from ...agents.run_config import StreamingMode
 from ...agents.transcription_entry import TranscriptionEntry
 from ...events.event import Event
-from ...features import FeatureName
-from ...features import is_feature_enabled
 from ...models.base_llm_connection import BaseLlmConnection
 from ...models.llm_request import LlmRequest
 from ...models.llm_response import LlmResponse
@@ -274,6 +272,25 @@ async def _send_to_model(
         await llm_connection.send_realtime(live_request.blob)
 
       if live_request.content:
+        content = live_request.content
+        # Persist user text content to session (similar to non-live mode)
+        # Skip function responses - they are already handled separately
+        is_function_response = content.parts and any(
+            part.function_response for part in content.parts
+        )
+        if not is_function_response:
+          if not content.role:
+            content.role = 'user'
+          user_content_event = Event(
+              id=Event.new_id(),
+              invocation_id=invocation_context.invocation_id,
+              author='user',
+              content=content,
+          )
+          await invocation_context.session_service.append_event(
+              session=invocation_context.session,
+              event=user_content_event,
+          )
         await llm_connection.send_content(live_request.content)
 
   async def _receive_from_model(
@@ -393,8 +410,8 @@ async def _run_one_step_async(
         current_invocation=True, current_branch=True
     )
 
-    # Long-running tool calls should have been handled before this point.
-    # If there are still long-running tool calls, it means the agent is paused
+    # Long running tool calls should have been handled before this point.
+    # If there are still long running tool calls, it means the agent is paused
     # before, and its branch hasn't been resumed yet.
     if (
         invocation_context.is_resumable
@@ -551,14 +568,11 @@ async def _postprocess_async(
     # Handles function calls.
     if model_response_event.get_function_calls():
 
-      if is_feature_enabled(FeatureName.PROGRESSIVE_SSE_STREAMING):
-        # In progressive SSE streaming mode stage 1, we skip partial FC events
-        # Only execute FCs in the final aggregated event (partial=False)
-        if (
-            invocation_context.run_config.streaming_mode == StreamingMode.SSE
-            and model_response_event.partial
-        ):
-          return
+      # Skip partial function call events - they should not trigger execution
+      # since partial events are not saved to session (see runners.py).
+      # Only execute function calls in the non-partial events.
+      if model_response_event.partial:
+        return
 
       async with Aclosing(
           self._postprocess_handle_function_calls_async(

src/google/adk/runners.py

@@ -815,6 +815,7 @@ async def _exec_with_plugin(
                   await self.session_service.append_event(
                       session=session, event=buffered_event
                   )
+                  yield buffered_event  # yield buffered events to caller
                 buffered_events = []
               else:
                 # non-transcription event or empty transcription event, for