Implementation:Langchain ai Langgraph ManagedValue Base

Attribute	Value
Source	`libs/langgraph/langgraph/managed/base.py` (31 lines)
Domain	Managed_Values, Core
Principle	Managed_Value_System
Library	langgraph
Import	`from langgraph.managed.base import ManagedValue, ManagedValueSpec`

Overview

The `base.py` module defines the abstract base class for LangGraph's managed value system. `ManagedValue` is a generic ABC that provides a static `get` method for computing values from the internal Pregel scratchpad. This module also provides type aliases and a type guard function for working with managed value specifications.

Description

Managed values are a mechanism that allows LangGraph to inject computed, runtime-scoped values into graph node signatures. Rather than storing these values in the graph state, managed values are computed on-the-fly from the execution scratchpad each time a node is invoked.

`ManagedValue(ABC, Generic[V])` -- The abstract base class for all managed values. It declares a single abstract static method:

• `get(scratchpad: PregelScratchpad) -> V` -- Computes and returns the managed value from the current execution scratchpad.

`ManagedValueSpec` -- A type alias equal to `type[ManagedValue]`. Used to annotate specifications that identify a managed value class.

`is_managed_value(value: Any) -> TypeGuard[ManagedValueSpec]` -- A type guard function that returns `True` if the given value is a class that is a subclass of `ManagedValue`. This is used internally to detect managed value annotations in graph state schemas.

`ManagedValueMapping` -- A type alias equal to `dict[str, ManagedValueSpec]`. Represents a mapping from state key names to their managed value classes.

The `PregelScratchpad` provides execution metadata such as the current step number and the maximum step limit, which managed value implementations (like `IsLastStepManager` and `RemainingStepsManager`) use to compute their values.

Usage

from langgraph.managed.base import ManagedValue, ManagedValueSpec, is_managed_value

# Check if a class is a managed value
class MyManager(ManagedValue[str]):
    @staticmethod
    def get(scratchpad):
        return "computed_value"

assert is_managed_value(MyManager)  # True
assert not is_managed_value(str)     # False

Code Reference

ManagedValue Class

Member	Signature	Description
`get`	`@staticmethod @abstractmethod get(scratchpad: PregelScratchpad) -> V`	Computes and returns the managed value from the current execution scratchpad.

Type Aliases

Name	Type	Description
`ManagedValueSpec`	`type[ManagedValue]`	A managed value class reference.
`ManagedValueMapping`	`dict[str, ManagedValueSpec]`	Maps state key names to their managed value classes.

Type Guard

Function	Signature	Description
`is_managed_value`	`(value: Any) -> TypeGuard[ManagedValueSpec]`	Returns `True` if `value` is a class that is a subclass of `ManagedValue`.

I/O Contract

Aspect	Detail
Input	`ManagedValue.get` receives a `PregelScratchpad` containing execution metadata (step number, stop limit, etc.).
Output	Returns a value of generic type `V` computed from the scratchpad.
Side Effects	None. Managed values are pure computations from the scratchpad state.

Usage Examples

Creating a Custom Managed Value

from typing import Annotated
from langgraph.managed.base import ManagedValue
from langgraph._internal._scratchpad import PregelScratchpad

class CurrentStepManager(ManagedValue[int]):
    @staticmethod
    def get(scratchpad: PregelScratchpad) -> int:
        return scratchpad.step

# Use with Annotated to inject into node signatures
CurrentStep = Annotated[int, CurrentStepManager]

Using is_managed_value for Introspection

from langgraph.managed.base import is_managed_value, ManagedValue

class MyManager(ManagedValue[str]):
    @staticmethod
    def get(scratchpad):
        return "hello"

print(is_managed_value(MyManager))  # True
print(is_managed_value(int))         # False

Related Pages

• Langchain_ai_Langgraph_IsLastStep -- Concrete managed value implementations built on this base class.
• Langchain_ai_Langgraph_Typing_Module -- Type variables used across LangGraph.
• Langchain_ai_Langgraph_PregelProtocol -- The Pregel execution protocol that provides the scratchpad.