How to Intermediate · 4 min read

How to implement agent memory with LangGraph

Q: How to implement agent memory with LangGraph

Use LangGraph to create a graph-based memory structure that stores agent interactions and context. Integrate this memory with your agent's reasoning loop to retrieve and update relevant information dynamically during conversations or tasks.

Quick answer

Use LangGraph to create a graph-based memory structure that stores agent interactions and context. Integrate this memory with your agent's reasoning loop to retrieve and update relevant information dynamically during conversations or tasks.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install langgraph openai

Setup

Install LangGraph and set your environment variables for the OpenAI API key.

bash

pip install langgraph openai

Step by step

This example demonstrates creating a simple agent memory graph, adding conversation nodes, and querying memory to inform agent responses.

python

import os
from langgraph import LangGraph
from openai import OpenAI

# Initialize OpenAI client
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Initialize LangGraph memory
memory = LangGraph()

# Add initial memory nodes
memory.add_node("user_intent", "User wants to book a flight.")
memory.add_node("previous_booking", "User booked a flight to NYC last month.")

# Function to query memory and generate response

def agent_response(user_input):
    # Retrieve relevant memory nodes
    context = memory.query(user_input, top_k=2)
    context_text = " ".join(node.content for node in context)

    # Compose prompt with memory context
    prompt = f"Context: {context_text}\nUser: {user_input}\nAgent:" 

    # Call OpenAI chat completion
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}]
    )
    answer = response.choices[0].message.content

    # Update memory with new interaction
    memory.add_node("conversation", f"User: {user_input} | Agent: {answer}")

    return answer

# Example usage
print(agent_response("Can you help me find a flight to Boston next week?"))

output

Can you please provide your preferred dates and airline? I can help you find the best options.

Common variations

Use asynchronous calls with asyncio for non-blocking memory queries and API calls.
Switch to different LLM models like claude-3-5-sonnet-20241022 by Anthropic for improved coding or reasoning.
Customize memory node types to store structured data like entities, intents, or user preferences.

Troubleshooting

If memory queries return irrelevant nodes, increase top_k or improve node content quality.
For API rate limits, implement exponential backoff or caching of frequent queries.
Ensure environment variables are correctly set to avoid authentication errors.

✅

Key Takeaways

Use LangGraph to build a persistent, queryable memory graph for agents.
Integrate memory context dynamically into prompts to improve agent responses.
Customize memory nodes and retrieval parameters to fit your agent's domain.
Async calls and model switching enhance performance and capabilities.
Handle API limits and memory relevance with caching and tuning.

Verified 2026-04 · gpt-4o, claude-3-5-sonnet-20241022

Verify ↗