langgraph-prebuilt 0.1.2__tar.gz → 0.1.3__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/PKG-INFO +1 -1
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/chat_agent_executor.py +67 -62
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/pyproject.toml +1 -1
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/LICENSE +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/README.md +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/__init__.py +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/interrupt.py +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/py.typed +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/tool_node.py +0 -0
- {langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/tool_validator.py +0 -0
{langgraph_prebuilt-0.1.2 → langgraph_prebuilt-0.1.3}/langgraph/prebuilt/chat_agent_executor.py
RENAMED
|
@@ -10,6 +10,7 @@ from typing import (
|
|
|
10
10
|
TypeVar,
|
|
11
11
|
Union,
|
|
12
12
|
cast,
|
|
13
|
+
get_type_hints,
|
|
13
14
|
)
|
|
14
15
|
|
|
15
16
|
from langchain_core.language_models import (
|
|
@@ -57,13 +58,27 @@ class AgentState(TypedDict):
|
|
|
57
58
|
remaining_steps: RemainingSteps
|
|
58
59
|
|
|
59
60
|
|
|
61
|
+
class AgentStatePydantic(BaseModel):
|
|
62
|
+
"""The state of the agent."""
|
|
63
|
+
|
|
64
|
+
messages: Annotated[Sequence[BaseMessage], add_messages]
|
|
65
|
+
|
|
66
|
+
remaining_steps: RemainingSteps = 25
|
|
67
|
+
|
|
68
|
+
|
|
60
69
|
class AgentStateWithStructuredResponse(AgentState):
|
|
61
70
|
"""The state of the agent with a structured response."""
|
|
62
71
|
|
|
63
72
|
structured_response: StructuredResponse
|
|
64
73
|
|
|
65
74
|
|
|
66
|
-
|
|
75
|
+
class AgentStateWithStructuredResponsePydantic(AgentStatePydantic):
|
|
76
|
+
"""The state of the agent with a structured response."""
|
|
77
|
+
|
|
78
|
+
structured_response: StructuredResponse
|
|
79
|
+
|
|
80
|
+
|
|
81
|
+
StateSchema = TypeVar("StateSchema", bound=Union[AgentState, AgentStatePydantic])
|
|
67
82
|
StateSchemaType = Type[StateSchema]
|
|
68
83
|
|
|
69
84
|
PROMPT_RUNNABLE_NAME = "Prompt"
|
|
@@ -76,21 +91,29 @@ Prompt = Union[
|
|
|
76
91
|
]
|
|
77
92
|
|
|
78
93
|
|
|
94
|
+
def _get_state_value(state: StateSchema, key: str, default: Any = None) -> Any:
|
|
95
|
+
return (
|
|
96
|
+
state.get(key, default)
|
|
97
|
+
if isinstance(state, dict)
|
|
98
|
+
else getattr(state, key, default)
|
|
99
|
+
)
|
|
100
|
+
|
|
101
|
+
|
|
79
102
|
def _get_prompt_runnable(prompt: Optional[Prompt]) -> Runnable:
|
|
80
103
|
prompt_runnable: Runnable
|
|
81
104
|
if prompt is None:
|
|
82
105
|
prompt_runnable = RunnableCallable(
|
|
83
|
-
lambda state: state
|
|
106
|
+
lambda state: _get_state_value(state, "messages"), name=PROMPT_RUNNABLE_NAME
|
|
84
107
|
)
|
|
85
108
|
elif isinstance(prompt, str):
|
|
86
109
|
_system_message: BaseMessage = SystemMessage(content=prompt)
|
|
87
110
|
prompt_runnable = RunnableCallable(
|
|
88
|
-
lambda state: [_system_message] + state
|
|
111
|
+
lambda state: [_system_message] + _get_state_value(state, "messages"),
|
|
89
112
|
name=PROMPT_RUNNABLE_NAME,
|
|
90
113
|
)
|
|
91
114
|
elif isinstance(prompt, SystemMessage):
|
|
92
115
|
prompt_runnable = RunnableCallable(
|
|
93
|
-
lambda state: [prompt] + state
|
|
116
|
+
lambda state: [prompt] + _get_state_value(state, "messages"),
|
|
94
117
|
name=PROMPT_RUNNABLE_NAME,
|
|
95
118
|
)
|
|
96
119
|
elif inspect.iscoroutinefunction(prompt):
|
|
@@ -234,7 +257,7 @@ def _validate_chat_history(
|
|
|
234
257
|
@_convert_modifier_to_prompt
|
|
235
258
|
def create_react_agent(
|
|
236
259
|
model: Union[str, LanguageModelLike],
|
|
237
|
-
tools: Union[Sequence[BaseTool], ToolNode],
|
|
260
|
+
tools: Union[Sequence[Union[BaseTool, Callable]], ToolNode],
|
|
238
261
|
*,
|
|
239
262
|
prompt: Optional[Prompt] = None,
|
|
240
263
|
response_format: Optional[
|
|
@@ -283,7 +306,7 @@ def create_react_agent(
|
|
|
283
306
|
The graph will make a separate call to the LLM to generate the structured response after the agent loop is finished.
|
|
284
307
|
This is not the only strategy to get structured responses, see more options in [this guide](https://langchain-ai.github.io/langgraph/how-tos/react-agent-structured-output/).
|
|
285
308
|
state_schema: An optional state schema that defines graph state.
|
|
286
|
-
Must have `messages` and `
|
|
309
|
+
Must have `messages` and `remaining_steps` keys.
|
|
287
310
|
Defaults to `AgentState` that defines those two keys.
|
|
288
311
|
config_schema: An optional schema for configuration.
|
|
289
312
|
Use this to expose configurable parameters via agent.config_specs.
|
|
@@ -595,7 +618,8 @@ def create_react_agent(
|
|
|
595
618
|
if response_format is not None:
|
|
596
619
|
required_keys.add("structured_response")
|
|
597
620
|
|
|
598
|
-
|
|
621
|
+
schema_keys = set(get_type_hints(state_schema))
|
|
622
|
+
if missing_keys := required_keys - set(schema_keys):
|
|
599
623
|
raise ValueError(f"Missing required key(s) {missing_keys} in state_schema")
|
|
600
624
|
|
|
601
625
|
if state_schema is None:
|
|
@@ -636,35 +660,34 @@ def create_react_agent(
|
|
|
636
660
|
# our graph needs to check if these were called
|
|
637
661
|
should_return_direct = {t.name for t in tool_classes if t.return_direct}
|
|
638
662
|
|
|
639
|
-
|
|
640
|
-
def call_model(state: AgentState, config: RunnableConfig) -> AgentState:
|
|
641
|
-
_validate_chat_history(state["messages"])
|
|
642
|
-
response = cast(AIMessage, model_runnable.invoke(state, config))
|
|
643
|
-
# add agent name to the AIMessage
|
|
644
|
-
response.name = name
|
|
663
|
+
def _are_more_steps_needed(state: StateSchema, response: BaseMessage) -> bool:
|
|
645
664
|
has_tool_calls = isinstance(response, AIMessage) and response.tool_calls
|
|
646
665
|
all_tools_return_direct = (
|
|
647
666
|
all(call["name"] in should_return_direct for call in response.tool_calls)
|
|
648
667
|
if isinstance(response, AIMessage)
|
|
649
668
|
else False
|
|
650
669
|
)
|
|
651
|
-
|
|
652
|
-
|
|
653
|
-
|
|
654
|
-
|
|
655
|
-
and has_tool_calls
|
|
656
|
-
)
|
|
670
|
+
remaining_steps = _get_state_value(state, "remaining_steps", None)
|
|
671
|
+
is_last_step = _get_state_value(state, "is_last_step", False)
|
|
672
|
+
return (
|
|
673
|
+
(remaining_steps is None and is_last_step and has_tool_calls)
|
|
657
674
|
or (
|
|
658
|
-
|
|
659
|
-
and
|
|
675
|
+
remaining_steps is not None
|
|
676
|
+
and remaining_steps < 1
|
|
660
677
|
and all_tools_return_direct
|
|
661
678
|
)
|
|
662
|
-
or (
|
|
663
|
-
|
|
664
|
-
|
|
665
|
-
|
|
666
|
-
|
|
667
|
-
)
|
|
679
|
+
or (remaining_steps is not None and remaining_steps < 2 and has_tool_calls)
|
|
680
|
+
)
|
|
681
|
+
|
|
682
|
+
# Define the function that calls the model
|
|
683
|
+
def call_model(state: StateSchema, config: RunnableConfig) -> StateSchema:
|
|
684
|
+
messages = _get_state_value(state, "messages")
|
|
685
|
+
_validate_chat_history(messages)
|
|
686
|
+
response = cast(AIMessage, model_runnable.invoke(state, config))
|
|
687
|
+
# add agent name to the AIMessage
|
|
688
|
+
response.name = name
|
|
689
|
+
|
|
690
|
+
if _are_more_steps_needed(state, response):
|
|
668
691
|
return {
|
|
669
692
|
"messages": [
|
|
670
693
|
AIMessage(
|
|
@@ -676,34 +699,13 @@ def create_react_agent(
|
|
|
676
699
|
# We return a list, because this will get added to the existing list
|
|
677
700
|
return {"messages": [response]}
|
|
678
701
|
|
|
679
|
-
async def acall_model(state:
|
|
680
|
-
|
|
702
|
+
async def acall_model(state: StateSchema, config: RunnableConfig) -> StateSchema:
|
|
703
|
+
messages = _get_state_value(state, "messages")
|
|
704
|
+
_validate_chat_history(messages)
|
|
681
705
|
response = cast(AIMessage, await model_runnable.ainvoke(state, config))
|
|
682
706
|
# add agent name to the AIMessage
|
|
683
707
|
response.name = name
|
|
684
|
-
|
|
685
|
-
all_tools_return_direct = (
|
|
686
|
-
all(call["name"] in should_return_direct for call in response.tool_calls)
|
|
687
|
-
if isinstance(response, AIMessage)
|
|
688
|
-
else False
|
|
689
|
-
)
|
|
690
|
-
if (
|
|
691
|
-
(
|
|
692
|
-
"remaining_steps" not in state
|
|
693
|
-
and state.get("is_last_step", False)
|
|
694
|
-
and has_tool_calls
|
|
695
|
-
)
|
|
696
|
-
or (
|
|
697
|
-
"remaining_steps" in state
|
|
698
|
-
and state["remaining_steps"] < 1
|
|
699
|
-
and all_tools_return_direct
|
|
700
|
-
)
|
|
701
|
-
or (
|
|
702
|
-
"remaining_steps" in state
|
|
703
|
-
and state["remaining_steps"] < 2
|
|
704
|
-
and has_tool_calls
|
|
705
|
-
)
|
|
706
|
-
):
|
|
708
|
+
if _are_more_steps_needed(state, response):
|
|
707
709
|
return {
|
|
708
710
|
"messages": [
|
|
709
711
|
AIMessage(
|
|
@@ -716,11 +718,11 @@ def create_react_agent(
|
|
|
716
718
|
return {"messages": [response]}
|
|
717
719
|
|
|
718
720
|
def generate_structured_response(
|
|
719
|
-
state:
|
|
720
|
-
) ->
|
|
721
|
+
state: StateSchema, config: RunnableConfig
|
|
722
|
+
) -> StateSchema:
|
|
721
723
|
# NOTE: we exclude the last message because there is enough information
|
|
722
724
|
# for the LLM to generate the structured response
|
|
723
|
-
messages = state
|
|
725
|
+
messages = _get_state_value(state, "messages")[:-1]
|
|
724
726
|
structured_response_schema = response_format
|
|
725
727
|
if isinstance(response_format, tuple):
|
|
726
728
|
system_prompt, structured_response_schema = response_format
|
|
@@ -733,11 +735,11 @@ def create_react_agent(
|
|
|
733
735
|
return {"structured_response": response}
|
|
734
736
|
|
|
735
737
|
async def agenerate_structured_response(
|
|
736
|
-
state:
|
|
737
|
-
) ->
|
|
738
|
+
state: StateSchema, config: RunnableConfig
|
|
739
|
+
) -> StateSchema:
|
|
738
740
|
# NOTE: we exclude the last message because there is enough information
|
|
739
741
|
# for the LLM to generate the structured response
|
|
740
|
-
messages = state
|
|
742
|
+
messages = _get_state_value(state, "messages")[:-1]
|
|
741
743
|
structured_response_schema = response_format
|
|
742
744
|
if isinstance(response_format, tuple):
|
|
743
745
|
system_prompt, structured_response_schema = response_format
|
|
@@ -773,8 +775,8 @@ def create_react_agent(
|
|
|
773
775
|
)
|
|
774
776
|
|
|
775
777
|
# Define the function that determines whether to continue or not
|
|
776
|
-
def should_continue(state:
|
|
777
|
-
messages = state
|
|
778
|
+
def should_continue(state: StateSchema) -> Union[str, list]:
|
|
779
|
+
messages = _get_state_value(state, "messages")
|
|
778
780
|
last_message = messages[-1]
|
|
779
781
|
# If there is no function call, then we finish
|
|
780
782
|
if not isinstance(last_message, AIMessage) or not last_message.tool_calls:
|
|
@@ -824,8 +826,8 @@ def create_react_agent(
|
|
|
824
826
|
path_map=should_continue_destinations,
|
|
825
827
|
)
|
|
826
828
|
|
|
827
|
-
def route_tool_responses(state:
|
|
828
|
-
for m in reversed(state
|
|
829
|
+
def route_tool_responses(state: StateSchema) -> Literal["agent", "__end__"]:
|
|
830
|
+
for m in reversed(_get_state_value(state, "messages")):
|
|
829
831
|
if not isinstance(m, ToolMessage):
|
|
830
832
|
break
|
|
831
833
|
if m.name in should_return_direct:
|
|
@@ -857,4 +859,7 @@ __all__ = [
|
|
|
857
859
|
"create_react_agent",
|
|
858
860
|
"create_tool_calling_executor",
|
|
859
861
|
"AgentState",
|
|
862
|
+
"AgentStatePydantic",
|
|
863
|
+
"AgentStateWithStructuredResponse",
|
|
864
|
+
"AgentStateWithStructuredResponsePydantic",
|
|
860
865
|
]
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|