Quickstart: OpenAI Agents Session Management

Coming Soon

This tutorial is under development. Check back soon!

What You'll Build

An OpenAI agent with persistent conversation memory that:

• Remembers context — Previous messages persist across requests
• Survives restarts — Memory stored in Dapr state, not in-process
• Scales horizontally — Any instance can serve any session

Prerequisites

• Dapr CLI installed and initialized
• Python 3.11+
• An OpenAI API key

Architecture

┌──────────────┐         ┌──────────────┐
│   Request    │         │   Request    │
│  Session: A  │         │  Session: A  │
└──────┬───────┘         └──────┬───────┘
       │                        │
       ▼                        ▼
┌──────────────┐         ┌──────────────┐
│   OpenAI     │         │   OpenAI     │
│  Instance 1  │         │  Instance 2  │
└──────┬───────┘         └──────┬───────┘
       │                        │
       └────────┬───────────────┘
                ▼
       ┌────────────────┐
       │  Dapr State    │
       │  (Session A)   │
       └────────────────┘

Both instances read/write to the same session state. The user gets consistent memory regardless of which instance handles the request.

Key Concepts

Concept	Description
Session ID	Unique identifier for a conversation thread
State Store	Dapr component that persists session data (Redis, PostgreSQL, etc.)
Memory Pattern	Store/retrieve conversation history before each LLM call
TTL	Optional expiration for inactive sessions