Does Langgraph sustain state variables across each graph run?

[ZohaibRamzan]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from fastapi import FastAPI
from langserve import add_routes
import uvicorn
from fastapi.responses import RedirectResponse
from build_langraph import o_app
from langchain_core.runnables import chain
from langchain.schema import (
    AIMessage,
    HumanMessage,
    SystemMessage
)
f_app = FastAPI()


@chain
def custom_chain(text):
    inputs = {"question": [HumanMessage(content=f"{text}")]}
    return o_app.invoke(inputs, {"configurable": {"thread_id": "10"}})



add_routes(f_app, custom_chain.with_types(input_type=str, ), path="/chat")

if __name__ == "__main__":
    uvicorn.run(f_app, host="0.0.0.0", port=8000)

Description

I am deploying a langgraph through langserve. When i made an invocation, the state variables gets updated. When i made the 2nd invocation, the state variables(updated during first invocation) gets destroyed. I am hitting the deployed graph through postman. If i want to sustain the state variables across our invocations. Like what was the state in previous run, I must know get them in next graph run.

System Info

langchain-nomic
langchain_community
tiktoken
langchainhub
chromadb
langchain
langgraph==0.0.44
tavily-python
gpt4all
python-dotenv
openai
bs4
langchain-openai
langserve
fastapi
uvicorn

[dosubot[bot]]

To sustain state variables across invocations in your Langgraph deployment with Langserve, you can implement a custom memory class that retains and updates state variables. Here's an example of how you could modify the memory handling to achieve persistence:

from typing import Any, Dict, List
from langchain_core.memory import BaseMemory

class PersistentMemory(BaseMemory):
    """Memory that retains state across invocations."""
    memories: Dict[str, Any] = dict()

    @property
    def memory_variables(self) -> List[str]:
        return list(self.memories.keys())

    def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, str]:
        return self.memories

    def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
        """Updates the memory with new context."""
        for key, value in outputs.items():
            self.memories[key] = value

    def clear(self) -> None:
        """Clears the memory."""
        self.memories.clear()

This PersistentMemory class updates the memories dictionary with new context from outputs, allowing state variables to be retained and updated across invocations. The clear method can be used to manually clear the memory when needed.

This approach directly addresses the issue of state variables being reset between invocations by ensuring that updates to the state are persisted. Implement this class and use it as the memory component in your Langgraph application to maintain state across different runs.

Details

Sources

• libs/langchain/langchain/memory/simple.py
• libs/langchain/langchain/chains/base.py
• libs/core/langchain_core/beta/runnables/context.py
• libs/cli/langchain_cli/project_template/app/server.py
• libs/core/langchain_core/runnables/configurable.py

About Dosu
This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

To continue the conversation, mention @dosu.

[ZohaibRamzan]

Can you tell me how to use this in Langgraph?

[dosubot[bot]]

To integrate the PersistentMemory class with your Langgraph setup using Langserve for maintaining state across runs, follow these steps:

1. Integration with Langgraph: Ensure Langgraph is configured to use your PersistentMemory class. This involves modifying Langgraph's initialization to instantiate and use your memory class with the relevant components.

2. Use in Langserve: Modify Langserve's initialization code to utilize PersistentMemory for session management or state persistence. This step is crucial for ensuring that the state is maintained across different runs.

3. Code Example:

# Assuming PersistentMemory is implemented as shown previously
memory_instance = PersistentMemory()
# Use this memory_instance in your Langgraph/Langserve setup

4. Compatibility and Testing: Make sure your PersistentMemory class is fully compatible with LangChain's memory management system. Test thoroughly in a development environment to confirm that state is correctly maintained without data loss.

This approach requires modifying the initialization and configuration of both Langgraph and Langserve to use the PersistentMemory instance. Unfortunately, specific configuration steps can vary based on your setup and the versions of Langgraph and Langserve you're using. For detailed guidance, refer to the documentation or source code of these projects, or consult with someone experienced in configuring these components.

Details

Sources

• libs/langchain/langchain/memory/init.py
• libs/langchain/langchain/schema/memory.py
• libs/core/langchain_core/memory.py
• libs/langchain/langchain/memory/vectorstore.py
• libs/experimental/langchain_experimental/generative_agents/memory.py

About Dosu
This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

To continue the conversation, mention @dosu.

[ZohaibRamzan]

@hwchase17 kindly give me direction here!

[hinthornw]

https://langchain-ai.github.io/langgraph/tutorials/introduction/#part-3-adding-memory-to-the-chatbot
https://langchain-ai.github.io/langgraph/reference/checkpoints/
https://langchain-ai.github.io/langgraph/how-tos/persistence/

[samosun]

@hinthornw Hi, Based on my understanding, the memory persistence is generated after the invoke. However, for example, if I want to persistence some file data in the state beforehand and then directly use this cached file data to answer questions during the first invoke, how can I achieve this? For instance, if I want to save my name in the memory beforehand.
If you have any suggestions, I would greatly appreciate it.

[mdfarhanali34]

Did you find out how it's done @samosun