How to beginner · 3 min read

How to persist agent memory between sessions

Q: How to persist agent memory between sessions

To persist agent memory between sessions, save the memory state externally (e.g., in a file, database, or vector store) and reload it when the agent restarts. This allows the agent to maintain context across sessions by restoring previous conversation history or embeddings.

Quick answer

To persist agent memory between sessions, save the memory state externally (e.g., in a file, database, or vector store) and reload it when the agent restarts. This allows the agent to maintain context across sessions by restoring previous conversation history or embeddings.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the openai Python SDK and set your API key as an environment variable.

Run pip install openai to install the SDK.
Set your API key in your shell: export OPENAI_API_KEY='your_api_key' (Linux/macOS) or setx OPENAI_API_KEY "your_api_key" (Windows).

bash

pip install openai

Step by step

This example demonstrates saving and loading agent memory as a JSON file to persist conversation history between sessions.

python

import os
import json
from openai import OpenAI

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

# File to persist memory
MEMORY_FILE = "agent_memory.json"

# Load memory from file if exists
if os.path.exists(MEMORY_FILE):
    with open(MEMORY_FILE, "r") as f:
        memory = json.load(f)
else:
    memory = []  # start with empty memory

# Append new user message
user_message = "Hello, how are you?"
memory.append({"role": "user", "content": user_message})

# Call the chat completion with memory as context
response = client.chat.completions.create(
    model="gpt-4o",
    messages=memory
)

# Append assistant response to memory
assistant_message = response.choices[0].message.content
memory.append({"role": "assistant", "content": assistant_message})

print("Assistant:", assistant_message)

# Persist updated memory to file
with open(MEMORY_FILE, "w") as f:
    json.dump(memory, f, indent=2)

output

Assistant: I'm doing well, thank you! How can I assist you today?

Common variations

You can persist memory using different storage backends:

Databases: Store conversation history in SQL or NoSQL databases for scalability.
Vector stores: Save embeddings for semantic search and retrieval (e.g., FAISS, ChromaDB).
Async usage: Use async SDK methods to load/save memory without blocking.
Different models: Use claude-3-5-sonnet-20241022 or gemini-1.5-pro with the same memory persistence approach.

Troubleshooting

If memory file is corrupted or unreadable, delete it to reset memory.
If context length exceeds model limits, truncate older messages before saving.
Ensure environment variable OPENAI_API_KEY is set correctly to avoid authentication errors.

✅

Key Takeaways

Persist agent memory externally to maintain context across sessions.
Use JSON files, databases, or vector stores depending on scale and use case.
Always reload and update memory before and after each interaction.
Truncate memory if it exceeds model token limits to avoid errors.
Set environment variables securely for API authentication.

Verified 2026-04 · gpt-4o, claude-3-5-sonnet-20241022, gemini-1.5-pro

Verify ↗