The agent engineering platform.
Python agentsaiai-agentsanthropicchatgptdeepagentsenterpriseframeworkgeminigenerative-ailangchainlanggraphllmmultiagentopen-sourceopenaipydanticpythonragtypescript
fix(langchain_v1): injection logic in tool node (#33344)
Eugene Yurtsev6a2efd060e0a
diff --git a/libs/langchain_v1/langchain/agents/factory.py b/libs/langchain_v1/langchain/agents/factory.pyindex 4b4ea63787..fc1b67ca76 100644--- a/libs/langchain_v1/langchain/agents/factory.py+++ b/libs/langchain_v1/langchain/agents/factory.py@@ -44,6 +44,7 @@ from langchain.agents.structured_output import ()from langchain.chat_models import init_chat_modelfrom langchain.tools import ToolNode+from langchain.tools.tool_node import ToolCallWithContextif TYPE_CHECKING:from collections.abc import Callable, Sequence@@ -865,7 +866,6 @@ def create_agent( # noqa: PLR0915_make_model_to_tools_edge(model_destination=loop_entry_node,structured_output_tools=structured_output_tools,- tool_node=tool_node,end_destination=exit_node,),model_to_tools_destinations,@@ -1023,11 +1023,11 @@ def _make_model_to_tools_edge(*,model_destination: str,structured_output_tools: dict[str, OutputToolBinding],- tool_node: ToolNode,end_destination: str,) -> Callable[[dict[str, Any], Runtime[ContextT]], str | list[Send] | None]:def model_to_tools(- state: dict[str, Any], runtime: Runtime[ContextT]+ state: dict[str, Any],+ runtime: Runtime[ContextT], # noqa: ARG001) -> str | list[Send] | None:# 1. if there's an explicit jump_to in the state, use itif jump_to := state.get("jump_to"):@@ -1053,11 +1053,17 @@ def _make_model_to_tools_edge(# 3. if there are pending tool calls, jump to the tool nodeif pending_tool_calls:- pending_tool_calls = [- tool_node.inject_tool_args(call, state, runtime.store)- for call in pending_tool_calls+ return [+ Send(+ "tools",+ ToolCallWithContext(+ __type="tool_call_with_context",+ tool_call=tool_call,+ state=state,+ ),+ )+ for tool_call in pending_tool_calls]- return [Send("tools", [tool_call]) for tool_call in pending_tool_calls]# 4. if there is a structured response, exit the loopif "structured_response" in state:diff --git a/libs/langchain_v1/langchain/tools/tool_node.py b/libs/langchain_v1/langchain/tools/tool_node.pyindex 2c1df81cf2..f0bbb1754c 100644--- a/libs/langchain_v1/langchain/tools/tool_node.py+++ b/libs/langchain_v1/langchain/tools/tool_node.py@@ -47,6 +47,7 @@ from typing import (Any,Literal,Optional,+ TypedDict,Union,cast,get_args,@@ -100,6 +101,29 @@ TOOL_INVOCATION_ERROR_TEMPLATE = ()+class ToolCallWithContext(TypedDict):+ """ToolCall with additional context for graph state.++ This is an internal data structure meant to help the ToolNode accept+ tool calls with additional context (e.g. state) when dispatched using the+ Send API.++ The Send API is used in create_agent to distribute tool calls in parallel+ and support human-in-the-loop workflows where graph execution may be paused+ for an indefinite time.+ """++ tool_call: ToolCall+ __type: Literal["tool_call_with_context"]+ """Type to parameterize the payload.++ Using "__" as a prefix to be defensive against potential name collisions with+ regular user state.+ """+ state: Any+ """The state is provided as additional context."""++def msg_content_output(output: Any) -> str | list[dict]:"""Convert tool output to valid message content format.@@ -431,7 +455,8 @@ class ToolNode(RunnableCallable):*,store: Optional[BaseStore], # noqa: UP045) -> Any:- tool_calls, input_type = self._parse_input(input, store)+ tool_calls, input_type = self._parse_input(input)+ tool_calls = [self._inject_tool_args(call, input, store) for call in tool_calls]config_list = get_config_list(config, len(tool_calls))input_types = [input_type] * len(tool_calls)with get_executor_for_config(config) as executor:@@ -446,7 +471,8 @@ class ToolNode(RunnableCallable):*,store: Optional[BaseStore], # noqa: UP045) -> Any:- tool_calls, input_type = self._parse_input(input, store)+ tool_calls, input_type = self._parse_input(input)+ tool_calls = [self._inject_tool_args(call, input, store) for call in tool_calls]outputs = await asyncio.gather(*(self._arun_one(call, input_type, config) for call in tool_calls))@@ -631,7 +657,6 @@ class ToolNode(RunnableCallable):def _parse_input(self,input: list[AnyMessage] | dict[str, Any] | BaseModel,- store: BaseStore | None,) -> tuple[list[ToolCall], Literal["list", "dict", "tool_calls"]]:input_type: Literal["list", "dict", "tool_calls"]if isinstance(input, list):@@ -641,6 +666,14 @@ class ToolNode(RunnableCallable):return tool_calls, input_typeinput_type = "list"messages = input+ elif isinstance(input, dict) and input.get("__type") == "tool_call_with_context":+ # Handle ToolCallWithContext from Send API+ # mypy will not be able to type narrow correctly since the signature+ # for input contains dict[str, Any]. We'd need to narrow dict[str, Any]+ # before we can apply correct typing.+ input_with_ctx = cast("ToolCallWithContext", input)+ input_type = "tool_calls"+ return [input_with_ctx["tool_call"]], input_typeelif isinstance(input, dict) and (messages := input.get(self._messages_key, [])):input_type = "dict"elif messages := getattr(input, self._messages_key, []):@@ -656,9 +689,7 @@ class ToolNode(RunnableCallable):msg = "No AIMessage found in input"raise ValueError(msg)- tool_calls = [- self.inject_tool_args(call, input, store) for call in latest_ai_message.tool_calls- ]+ tool_calls = list(latest_ai_message.tool_calls)return tool_calls, input_typedef _validate_tool_call(self, call: ToolCall) -> ToolMessage | None:@@ -696,14 +727,20 @@ class ToolNode(RunnableCallable):err_msg += f" State should contain fields {required_fields_str}."raise ValueError(err_msg)- if isinstance(input, dict):+ # Extract state from ToolCallWithContext if present+ if isinstance(input, dict) and input.get("__type") == "tool_call_with_context":+ state = input["state"]+ else:+ state = input++ if isinstance(state, dict):tool_state_args = {- tool_arg: input[state_field] if state_field else input+ tool_arg: state[state_field] if state_field else statefor tool_arg, state_field in state_args.items()}else:tool_state_args = {- tool_arg: getattr(input, state_field) if state_field else input+ tool_arg: getattr(state, state_field) if state_field else statefor tool_arg, state_field in state_args.items()}@@ -731,7 +768,7 @@ class ToolNode(RunnableCallable):}return tool_call- def inject_tool_args(+ def _inject_tool_args(self,tool_call: ToolCall,input: list[AnyMessage] | dict[str, Any] | BaseModel,@@ -739,10 +776,11 @@ class ToolNode(RunnableCallable):) -> ToolCall:"""Inject graph state and store into tool call arguments.- This method enables tools to access graph context that should not be controlled- by the model. Tools can declare dependencies on graph state or persistent storage- using InjectedState and InjectedStore annotations. This method automatically- identifies these dependencies and injects the appropriate values.+ This is an internal method that enables tools to access graph context that+ should not be controlled by the model. Tools can declare dependencies on graph+ state or persistent storage using InjectedState and InjectedStore annotations.+ This method automatically identifies these dependencies and injects the+ appropriate values.The injection process preserves the original tool call structure while addingthe necessary context arguments. This allows tools to be both model-callable@@ -765,10 +803,8 @@ class ToolNode(RunnableCallable):or if state injection requirements cannot be satisfied.Note:- This method is automatically called during tool execution but can also- be used manually when working with the Send API or custom routing logic.- The injection is performed on a copy of the tool call to avoid mutating- the original.+ This method is called automatically during tool execution. It should not+ be called from outside the ToolNode."""if tool_call["name"] not in self.tools_by_name:return tool_call