.Net: Feature: Mimir as a Semantic Kernel memory connector (offline, encrypted, single-binary, cross-platform)

Feature request: Mimir as a Semantic Kernel memory connector

Problem

Semantic Kernel's memory system (ISemanticTextMemory, IMemoryStore) currently supports vector DBs (Azure AI Search, Chroma, Pinecone, Qdrant, etc.) but lacks an offline, single-binary, zero-infrastructure option for local development, desktop apps, and edge deployments.

Proposed solution

Add Mimir as an IMemoryStore implementation — a single Rust binary (~8MB) with embedded SQLite + FTS5 + vector search + AES-256 encryption.

Mimir is an MIT-licensed, MCP-native persistent memory engine for AI agents. Single Rust binary (~8MB), zero runtime dependencies.

Key features:

• 29 MCP tools — full memory lifecycle: remember, recall (FTS5), search (semantic), forget, decay, embed, cohere, vault export/import, journal, timeline
• Hybrid search — BM25 (FTS5) + dense embeddings + Reciprocal Rank Fusion
• AES-256-GCM encryption at rest
• Confidence decay — Ebbinghaus-based, auto-archives stale memories
• Web dashboard for browsing/editing memories
• Single binary — no Docker, Postgres, or cloud required
• MIT licensed — no restrictions

Why Mimir for Semantic Kernel

1. IMemoryStore interface maps cleanly to Mimir's CRUD operations
2. Hybrid search (BM25 + vectors) gives better retrieval than pure vector search alone
3. Offline-first — no Azure/AWS/GCP dependency. Works on dev laptops, factory floors, IoT
4. Encryption at rest — enterprise requirement that most other memory stores don't meet
5. MCP-native — Semantic Kernel supports MCP plugins; Mimir tools are immediately available
6. .NET-friendly — Mimir is invoked as a subprocess via MCP stdio, language-agnostic

Integration (Python example)

from semantic_kernel.memory import SemanticTextMemory
from mimir_memory_store import MimirMemoryStore  # IMemoryStore implementation

store = MimirMemoryStore(db_path="./kernel_memory.db")
memory = SemanticTextMemory(storage=store, embeddings_generator=embedding_service)

await memory.save_information("user", "prefers dark mode", "pref-001")
results = await memory.search("user", "mode preference")  # Survives restarts

Status

• Repo: Perseus-Computing-LLC/mimir
• License: MIT
• Cross-platform Rust binary (Linux, macOS, Windows)

Would a PR implementing Mimir as an IMemoryStore connector be welcome?