ADK changes

PiperOrigin-RevId: 814319961
This commit is contained in:
Xuan Yang
2025-10-02 13:44:05 -07:00
committed by Copybara-Service
parent 2e2d61b6fe
commit d3148dacc9
6 changed files with 622 additions and 6 deletions
+24 -2
View File
@@ -112,8 +112,22 @@ ToolUnion: TypeAlias = Union[Callable, BaseTool, BaseToolset]
async def _convert_tool_union_to_tools(
tool_union: ToolUnion, ctx: ReadonlyContext
tool_union: ToolUnion,
ctx: ReadonlyContext,
model: Union[str, BaseLlm],
multiple_tools: bool = False,
) -> list[BaseTool]:
from ..tools.google_search_tool import google_search
# Wrap google_search tool with AgentTool if there are multiple tools because
# the built-in tools cannot be used together with other tools.
# TODO(b/448114567): Remove once the workaround is no longer needed.
if multiple_tools and tool_union is google_search:
from ..tools.google_search_agent_tool import create_google_search_agent
from ..tools.google_search_agent_tool import GoogleSearchAgentTool
return [GoogleSearchAgentTool(create_google_search_agent(model))]
if isinstance(tool_union, BaseTool):
return [tool_union]
if callable(tool_union):
@@ -462,8 +476,16 @@ class LlmAgent(BaseAgent):
This method is only for use by Agent Development Kit.
"""
resolved_tools = []
# We may need to wrap some built-in tools if there are other tools
# because the built-in tools cannot be used together with other tools.
# TODO(b/448114567): Remove once the workaround is no longer needed.
multiple_tools = len(self.tools) > 1
for tool_union in self.tools:
resolved_tools.extend(await _convert_tool_union_to_tools(tool_union, ctx))
resolved_tools.extend(
await _convert_tool_union_to_tools(
tool_union, ctx, self.model, multiple_tools
)
)
return resolved_tools
@property
@@ -45,6 +45,7 @@ from ...telemetry.tracing import trace_call_llm
from ...telemetry.tracing import trace_send_data
from ...telemetry.tracing import tracer
from ...tools.base_toolset import BaseToolset
from ...tools.google_search_tool import google_search
from ...tools.tool_context import ToolContext
from ...utils.context_utils import Aclosing
from .audio_cache_manager import AudioCacheManager
@@ -442,6 +443,11 @@ class BaseLlmFlow(ABC):
yield event
# Run processors for tools.
# We may need to wrap some built-in tools if there are other tools
# because the built-in tools cannot be used together with other tools.
# TODO(b/448114567): Remove once the workaround is no longer needed.
multiple_tools = len(agent.tools) > 1
for tool_union in agent.tools:
tool_context = ToolContext(invocation_context)
@@ -455,7 +461,10 @@ class BaseLlmFlow(ABC):
# Then process all tools from this tool union
tools = await _convert_tool_union_to_tools(
tool_union, ReadonlyContext(invocation_context)
tool_union,
ReadonlyContext(invocation_context),
llm_request.model,
multiple_tools,
)
for tool in tools:
await tool.process_llm_request(
@@ -818,6 +827,26 @@ class BaseLlmFlow(ABC):
agent = invocation_context.agent
# Add grounding metadata to the response if needed.
# TODO(b/448114567): Remove this function once the workaround is no longer needed.
async def _maybe_add_grounding_metadata(
response: Optional[LlmResponse] = None,
) -> Optional[LlmResponse]:
readonly_context = ReadonlyContext(invocation_context)
tools = await agent.canonical_tools(readonly_context)
if not any(tool.name == 'google_search_agent' for tool in tools):
return response
ground_metadata = invocation_context.session.state.get(
'temp:_adk_grounding_metadata', None
)
if not ground_metadata:
return response
if not response:
response = llm_response
response.grounding_metadata = ground_metadata
return response
callback_context = CallbackContext(
invocation_context, event_actions=model_response_event.actions
)
@@ -830,12 +859,12 @@ class BaseLlmFlow(ABC):
)
)
if callback_response:
return callback_response
return await _maybe_add_grounding_metadata(callback_response)
# If no overrides are provided from the plugins, further run the canonical
# callbacks.
if not agent.canonical_after_model_callbacks:
return
return await _maybe_add_grounding_metadata()
for callback in agent.canonical_after_model_callbacks:
callback_response = callback(
callback_context=callback_context, llm_response=llm_response
@@ -843,7 +872,8 @@ class BaseLlmFlow(ABC):
if inspect.isawaitable(callback_response):
callback_response = await callback_response
if callback_response:
return callback_response
return await _maybe_add_grounding_metadata(callback_response)
return await _maybe_add_grounding_metadata()
def _finalize_model_response_event(
self,
@@ -0,0 +1,140 @@
# Copyright 2025 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import annotations
from typing import Any
from typing import Union
from google.genai import types
from typing_extensions import override
from ..agents.llm_agent import LlmAgent
from ..memory.in_memory_memory_service import InMemoryMemoryService
from ..models.base_llm import BaseLlm
from ..utils.context_utils import Aclosing
from ._forwarding_artifact_service import ForwardingArtifactService
from .agent_tool import AgentTool
from .google_search_tool import google_search
from .tool_context import ToolContext
def create_google_search_agent(model: Union[str, BaseLlm]) -> LlmAgent:
"""Create a sub-agent that only uses google_search tool."""
return LlmAgent(
name='google_search_agent',
model=model,
description=(
'An agent for performing Google search using the `google_search` tool'
),
instruction="""
You are a specialized Google search agent.
When given a search query, use the `google_search` tool to find the related information.
""",
tools=[google_search],
)
class GoogleSearchAgentTool(AgentTool):
"""A tool that wraps a sub-agent that only uses google_search tool.
This is a workaround to support using google_search tool with other tools.
TODO(b/448114567): Remove once the workaround is no longer needed.
Attributes:
model: The model to use for the sub-agent.
"""
def __init__(self, agent: LlmAgent):
self.agent = agent
super().__init__(agent=self.agent)
@override
async def run_async(
self,
*,
args: dict[str, Any],
tool_context: ToolContext,
) -> Any:
from ..agents.llm_agent import LlmAgent
from ..runners import Runner
from ..sessions.in_memory_session_service import InMemorySessionService
if isinstance(self.agent, LlmAgent) and self.agent.input_schema:
input_value = self.agent.input_schema.model_validate(args)
content = types.Content(
role='user',
parts=[
types.Part.from_text(
text=input_value.model_dump_json(exclude_none=True)
)
],
)
else:
content = types.Content(
role='user',
parts=[types.Part.from_text(text=args['request'])],
)
runner = Runner(
app_name=self.agent.name,
agent=self.agent,
artifact_service=ForwardingArtifactService(tool_context),
session_service=InMemorySessionService(),
memory_service=InMemoryMemoryService(),
credential_service=tool_context._invocation_context.credential_service,
plugins=list(tool_context._invocation_context.plugin_manager.plugins),
)
state_dict = {
k: v
for k, v in tool_context.state.to_dict().items()
if not k.startswith('_adk') # Filter out adk internal states
}
session = await runner.session_service.create_session(
app_name=self.agent.name,
user_id=tool_context._invocation_context.user_id,
state=state_dict,
)
last_content = None
last_grounding_metadata = None
async with Aclosing(
runner.run_async(
user_id=session.user_id, session_id=session.id, new_message=content
)
) as agen:
async for event in agen:
# Forward state delta to parent session.
if event.actions.state_delta:
tool_context.state.update(event.actions.state_delta)
if event.content:
last_content = event.content
last_grounding_metadata = event.grounding_metadata
if not last_content:
return ''
merged_text = '\n'.join(p.text for p in last_content.parts if p.text)
if isinstance(self.agent, LlmAgent) and self.agent.output_schema:
tool_result = self.agent.output_schema.model_validate_json(
merged_text
).model_dump(exclude_none=True)
else:
tool_result = merged_text
if last_grounding_metadata:
tool_context.state['temp:_adk_grounding_metadata'] = (
last_grounding_metadata
)
return tool_result