Microsoft Agent Framework + MemorySync

Overview

Install agent-framework-memorysync and Microsoft Agent Framework agents remember users across sessions. MemorySyncContextProvider rides the framework’s ContextProvider seam: before every run, relevant memories are recalled under a hard budget (default 1.2s) and injected into the instructions layer — the system prompt, where context belongs (the first-party Mem0 provider fabricates a role="user" message instead). After every run both sides of the exchange persist automatically with deterministic idempotency seeds, once per user turn even in tool loops, and entirely fail-open — the Mem0 provider’s after_run errors crash the agent after a successful run. One user_id drives storage AND retrieval (Mem0 splits them; forget the search scope and your agent is silently memoryless), and a per-run identity ladder — agent.run options, a resolver callable, then the constructor — solves the construction-bound identity limitation both Mem0 and Zep document. Python 3.10+, matching the framework itself (Zep requires 3.11).

Setup status and requirements

Supported version

agent-framework-memorysync 1.0.0 (PyPI)

Last setup review

2026-08-24

Permissions

A MemorySync API key with read and write scopes.

Limits

Requires agent-framework-core 1.8+ (Python 3.10+).

Capabilities

• Instructions-layer injection — never a fabricated user message
• Hard recall budget (default 1.2s) — a run is never stalled
• Capture is entirely fail-open — a memory outage can never crash the agent
• Per-run identity: agent.run options → resolver → constructor
• One user_id drives storage and retrieval — no silent memoryless agents
• once-per-turn capture in tool loops; deterministic idempotency seeds
• Session state stays JSON-native — AgentSession serialization keeps working
• Optional search_memory + save_memory tools via extend_tools

Quick Start

from agent_framework import Agent
from agent_framework_memorysync import MemorySyncContextProvider

provider = MemorySyncContextProvider(
    api_key="ms_...",
    user_id="customer-42",
    session_id="support",
)

agent = Agent(client=..., context_providers=[provider])
result = await agent.run("which seat should I book?")

# per-run identity (neither Mem0 nor Zep supports this):
await agent.run("...", options={"memorysync_user_id": "other-user"})

Use Cases

Explore More