langgraph/how-tos/state-model/

[giscus[bot]]

langgraph/how-tos/state-model/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/how-tos/state-model/

[whamsicore]

This is pretty cool, but i don't like how black box the model is. Can new tools be added to the model, post compile? Doesn't seem so. This is more of a defect of langchain, however

[eloko7]

You can place tool_executor = ToolExecutor(tools) and model = model.bind_tools(tools) inside a function that you invoke doing runtime. This function can also take config or state as input variables where you can hide some logic for changing or adding tools after compiling the graph.

[whamsicore]

That's very cool. Thx for the tip!

[whamsicore]

also, in the description of pydantic state it says:
"If you want to apply additional validation on state updates, you could instead opt for a pydantic BaseModel."
Does this example actually implement any additional validation on state updates? The only affect seems to be "messages attribute is treated as "append-only". How do I actually add validation on state change, then?

[rhlarora84]

Can you please help me understand other attributes when creating StateGraph?

StateGraph(AgentState, input=MessagesState, output=OutputState, config_schema=ConfigSchema)

• What is the difference between AgentState & input? Does the input reflects the input to the entrypoint and AgentState represents the state of the graph?
• Same applies for OutputState but before the END?

[rhlarora84]

Found the answer https://langchain-ai.github.io/langgraph/how-tos/input_output_schema/

[Sergears]

When I try to run this example, I get an error from the ToolNode: 'AgentState' object has no attribute 'get'. So seems like it still expects a TypedDict object for state (dependency versions langgraph=0.2.18, langchain-core=0.2.39)

[tomas-herman]

When I use BaseModel for the agent state and inside it I use a list of class also defined as BaseModel I am getting serialisation issues and "type":"not_implemented

Precise example is agent state

class AgentState(BaseModel):
"""State class for the chat agent.

This class uses Pydantic for type validation while maintaining compatibility
with LangGraph's message management system.

Attributes:
    user_id: Unique identifier for the user
    project_id: Unique identifier for the project
    messages: List of chat messages managed by LangGraph's add_messages function
    follow_up_questions: List of follow-up questions as strings
    references: List of Reference objects containing title and URL

"""

user_id: str = Field(..., description="Unique identifier for the user")
project_id: str = Field(..., description="Unique identifier for the project")
messages: Annotated[list[AnyMessage], add_messages] = Field(
    default_factory=list, description="List of chat messages managed by LangGraph"
)
follow_up_questions: list[str] = Field(
    default_factory=list, description="List of follow-up questions"
)
references: list[Reference] = Field(
    default_factory=list, description="List of references with title and URL"
)

And reference defnied as

class Reference(BaseModel):
"""Model representing a reference with a title and URL."""

title: str = Field(..., description="Title of the reference")
url: str = Field(..., description="URL link of the reference")

Are causing type not implemented

references":[{"lc":1,"type":"not_implemented","id":["app","state","Reference"],"repr":"Reference(title='How to Create a Successful Startup', url='https://example.com/startup-guide')"},{"lc":1,"type":"not_implemented","id":["app","state","Reference"],"repr":"Reference(title='Market Research Fundamentals', url='https://example.com/market-research')"},{"lc":1,"type":"not_implemented","id":["app","state","Reference"],"repr":"Reference(title='Startup Funding Strategies', url='https://example.com/funding-guide')"}]}"

While if I define the reference as TypedDict

class Reference(TypedDict):
"""Dictionary representing a reference with a title and URL."""

title: str  # Title of the reference
url: str  # URL link of the reference

All is fine

"references":[{"title":"How to Create a Successful Startup","url":"https://example.com/startup-guide"},{"title":"Market Research Fundamentals","url":"https://example.com/market-research"},{"title":"Startup Funding Strategies","url":"https://example.com/funding-guide"}]}"

Am I doing something wrong or why is the nested BaseModel not recognised correctly and the nested TypedDict works fine?

[Jackoder]

When I run python agent.py, everything is fine. But it throws exception when I run langgraph dev and submit with LangGraph Studio. What's the difference?

Traceback (most recent call last):
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph_api\queue.py", line 156, in worker
    await asyncio.wait_for(consume(stream, run_id), timeout)
  File "D:\AI\Anaconda3\envs\py311\Lib\asyncio\tasks.py", line 489, in wait_for
    return fut.result()
           ^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph_api\stream.py", line 285, in consume
    raise e from None
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph_api\stream.py", line 275, in consume
    async for mode, payload in stream:
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph_api\stream.py", line 226, in astream_state
    event = await wait_if_not_done(anext(stream, sentinel), done)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph_api\asyncio.py", line 67, in wait_if_not_done
    raise e.exceptions[0] from None
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\pregel\__init__.py", line 1874, in astream
    async for _ in runner.atick(
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\pregel\runner.py", line 444, in atick
    _panic_or_proceed(
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\pregel\runner.py", line 539, in _panic_or_proceed
    raise exc
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\pregel\retry.py", line 132, in arun_with_retry
    return await task.proc.ainvoke(task.input, config)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\utils\runnable.py", line 445, in ainvoke
    input = await step.ainvoke(input, config, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langgraph\utils\runnable.py", line 236, in ainvoke
    ret = await asyncio.create_task(coro, context=context)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\runnables\config.py", line 588, in run_in_executor
    return await asyncio.get_running_loop().run_in_executor(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\concurrent\futures\thread.py", line 58, in run
    result = self.fn(*self.args, **self.kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\runnables\config.py", line 579, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\llm-rag\./src/app.py", line 276, in grade_documents
    score = retrieval_grader.invoke(
            ^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\runnables\base.py", line 3024, in invoke
    input = context.run(step.invoke, input, config)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\output_parsers\base.py", line 193, in invoke
    return self._call_with_config(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\runnables\base.py", line 1927, in _call_with_config
    context.run(
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\runnables\config.py", line 396, in call_func_with_variable_args
    return func(input, **kwargs)  # type: ignore[call-arg]
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\output_parsers\base.py", line 194, in <lambda>
    lambda inner_input: self.parse_result(
                        ^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\output_parsers\openai_tools.py", line 298, in parse_result
    raise e
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\langchain_core\output_parsers\openai_tools.py", line 293, in parse_result
    pydantic_objects.append(name_dict[res["type"]](**res["args"]))
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\Anaconda3\envs\py311\Lib\site-packages\pydantic\main.py", line 193, in __init__
    self.__pydantic_validator__.validate_python(data, self_instance=self)
pydantic_core._pydantic_core.ValidationError: 1 validation error for GradeDocuments
binary_score
  Field required [type=missing, input_value={}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.8/v/missing
During task with name 'grade_documents' and id '3d60efe6-16a6-f1da-fca0-46ad323838ce'