.NET: Feature request: Mimir memory connector (local-first, encrypted, 23 MCP tools)

[tcconnally]

Feature Request: Mimir as a memory connector for Microsoft Agent Framework

Mimir is an open-source (MIT) persistent memory engine with 23 MCP tools — encrypted, local-first, embedded SQLite with hybrid search.

Proposal

Add a Mimir memory connector to the Microsoft Agent Framework's memory abstraction layer. This would complement the existing memory implementations and provide a zero-dependency, production-ready option for teams that can't or won't use cloud memory services.

Why this fits Agent Framework

• Agent Framework v1.0 has a modular memory connector architecture (.NET: Consolidate Memory implementations between Python and C# #6415 even calls for consolidating memory implementations between Python and C#)
• Mimir has both a Python client and a Rust core that compiles to a single binary — perfect for the "local-first" agent pattern
• The 23 MCP tools integrate naturally with the framework's MCP support (Python: Support MCP skills of mcp-resource-template type #6118 works on MCP skill integration)
• Encryption at rest addresses enterprise security requirements without adding cloud dependencies

Integration sketch

from agent_framework.memory import MemoryConnector
from mimir_client import MimirClient

class MimirMemoryConnector(MemoryConnector):
    def __init__(self, db_path: str):
        self.client = MimirClient(db_path)

    async def store(self, key: str, value: dict, metadata: dict = None):
        await self.client.remember(content=json.dumps(value), metadata=metadata)

    async def retrieve(self, query: str, limit: int = 10):
        results = await self.client.recall(query, limit=limit)
        return [r.content for r in results]

    async def forget(self, key: str):
        await self.client.forget(key)

Happy to contribute a PR.