OpenAI Agents SDK + MemorySync

Overview

Install openai-agents-memorysync and pass session=MemorySyncSession(...) to Runner.run: session items are kept server-side in MemorySync conversation history exactly as the SDK wrote them, survive restarts and deploys, and follow multi-agent handoffs. The transcript is never listed or recalled as memories; the durable facts in user messages go to long-term memory automatically, and assistant replies are not stored. memory_instructions adds a memory block to the system prompt each time the SDK builds it (by default the user’s newest facts, or recall against a prompt you supply), and five structured tools give agents explicit memory. No other memory vendor implements the SDK’s Session protocol.

Setup status and requirements

Supported version

openai-agents-memorysync 1.1.0 (PyPI)

Last setup review

2026-10-01

Permissions

A MemorySync API key with read and write scopes; pop_item/clear_session need delete permission.

Limits

Python 3.10+. Install openai-agents alongside it. Python only — TypeScript agents use the Vercel AI SDK or Mastra integrations.

Capabilities

• The first drop-in Session protocol implementation from any memory vendor
• History shared across multi-agent handoffs automatically
• Retry-safe transcript writes: total and partial batch failures converge
• Automatic long-term memory from the user’s messages
• pop_item/clear_session scoped and loud — never silent data loss
• memory_instructions: a memory block in the system prompt, with per-request identity
• Five never-raise agent tools with a read-only mode

Quick Start

from agents import Agent, Runner
from openai_agents_memorysync import MemorySyncSession

agent = Agent(name="Assistant", instructions="You are a helpful assistant.")
session = MemorySyncSession("thread-42", user_id="customer-7")

result = await Runner.run(agent, "Book my usual trip.", session=session)

Use Cases

Explore More