Building a Memory Agent with the OpenAI Agents SDK and Zep
A video walkthrough demonstrating using Zep's agent memory and the new OpenAI Agents SDK to build an AI agent with long-term memory.
I've been experimenting with the new OpenAI Agents SDK and built a memory agent using Zep. In this video, I'll walk you through a quick demo of how I've integrated the framework and Zep to create a memory agent CLI app, demonstrate the dynamic updating of user preferences, and showcase Zep's memory capabilities.
The new OpenAI Agents SDK
The new OpenAI Agents SDK provides a simple, nicely abstracted approach to build agentic applications without getting in your way. I found the SDK intuitive and lightweight, providing just enough structure to keep things clear and maintainable without becoming opaque or restrictive.
It boils down to three core concepts:
- Agents: LLMs equipped with instructions and tools.
- Handoffs: Easy delegation between agents for specialized tasks.
- Guardrails: Simple validation to ensure agent inputs stay on track.
These straightforward abstractions give you the flexibility to build complex, production-ready applications quickly. The framework has built-in tracing, making visualizing, debugging, and optimizing your agent simple.
This example memory agent is intentionally simple, with no hand-offs between agents and no guardrails.
If you've been searching for a solid agent framework but haven't found one that works for you, I'd recommend checking out the new OpenAI Agents SDK.
Next Steps
- Get the Agent source on GitHub
- Learn more about building with Zep
- Read the OpenAI SDK docs
- Sign up for a free Zep account.