AutoGen + MemorySync

Overview

Install autogen-memorysync and AutoGen agents remember users across sessions. MemorySyncMemory implements the full autogen_core Memory protocol: update_context runs before every model call and injects recalled context as a SystemMessage under a hard budget (default 1.2s) — a slow or dead memory backend means an unenriched turn, never a late or broken one. Retrieval is role-aware (the last user message, not blindly messages[-1] — which after a tool result is assistant text, a real retrieval bug in the Mem0 adapter that ships inside autogen-ext). Everything is genuinely async (Mem0’s calls a synchronous client inside async methods, blocking the event loop), writes carry deterministic idempotency seeds so retries never duplicate, clear() is session-scoped by default with whole-user wipe as an explicit opt-in, and add_turn_pair() reduces the capture loop AutoGen deliberately leaves to applications to one call. search_memory and save_memory FunctionTools included.

Setup status and requirements

Supported version

autogen-memorysync 1.0.0 (PyPI)

Last setup review

2026-08-24

Permissions

A MemorySync API key with read and write scopes.

Limits

Requires autogen-agentchat/autogen-core 0.4+ (Python 3.10+). AutoGen is in maintenance mode — the Memory protocol is frozen and stable; a Microsoft Agent Framework adapter is a separate roadmap item.

Capabilities

• Full autogen_core Memory protocol — drop into AssistantAgent memory=[...]
• Hard recall budget (default 1.2s) — a reply is never late, tested against a slow backend
• Genuinely async httpx — never blocks the event loop (Mem0’s adapter does)
• Role-aware retrieval: queries the last user message, not messages[-1]
• clear() is session-scoped by default — cannot wipe a customer by accident
• Deterministic idempotency seeds — retried adds converge on one row
• add_turn_pair(): the whole capture loop in one call
• search_memory + save_memory FunctionTools for tool-equipped agents

Quick Start

from autogen_agentchat.agents import AssistantAgent
from autogen_memorysync import MemorySyncMemory

memory = MemorySyncMemory(api_key="ms_...",
                          user_id="customer-42",
                          session_id="support")

agent = AssistantAgent("assistant", model_client=...,
                       memory=[memory])

result = await agent.run(task=user_text)
await memory.add_turn_pair(user_text,
                           result.messages[-1].content)

Use Cases

Explore More