Integration Tutorial April 13, 2026

Memory Plugin for Microsoft Semantic Kernel

Semantic Kernel's plugin architecture makes it straightforward to extend AI agents with new capabilities. This guide shows how to build a REM Labs memory plugin that gives your Semantic Kernel agents persistent, cross-session recall -- with vector, full-text, and entity graph retrieval built in.

Why Semantic Kernel Needs External Memory

Semantic Kernel provides a flexible orchestration layer for building AI agents in C# and Python. But its built-in memory connectors are limited to vector-only stores. For production use cases -- where you need to recall proper nouns, handle knowledge updates, and search across thousands of past interactions -- you need multi-signal retrieval. That is what REM Labs provides.

Step 1: Install

pip install remlabs-memory semantic-kernel

Step 2: Build the Memory Plugin

from semantic_kernel.functions import kernel_function
from remlabs import RemMemory

class RemMemoryPlugin:
    def __init__(self, api_key: str, namespace: str = "semantic-kernel"):
        self.mem = RemMemory(api_key=api_key)
        self.namespace = namespace

    @kernel_function(description="Search persistent memory for relevant context")
    def recall(self, query: str) -> str:
        results = self.mem.search(query, namespace=self.namespace, limit=5)
        if not results:
            return "No relevant memories found."
        return "\n".join([f"- {r['value']} (score: {r['score']:.2f})" for r in results])

    @kernel_function(description="Store a fact or observation in persistent memory")
    def remember(self, value: str, tags: str = "") -> str:
        tag_list = [t.strip() for t in tags.split(",") if t.strip()]
        self.mem.store(value=value, namespace=self.namespace, tags=tag_list)
        return f"Stored: {value}"

The plugin exposes two kernel functions: recall for semantic search and remember for storing new facts. Semantic Kernel's planner can invoke these automatically when the agent decides it needs context or wants to persist something.

Step 3: Register and Use

import semantic_kernel as sk
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion

kernel = sk.Kernel()

kernel.add_service(OpenAIChatCompletion(
    service_id="chat",
    ai_model_id="gpt-4o",
    api_key="..."
))

# Register the REM memory plugin
kernel.add_plugin(
    RemMemoryPlugin(api_key="sk-slop-..."),
    plugin_name="memory"
)

# The agent can now call memory.recall and memory.remember
result = await kernel.invoke_prompt(
    "What do you know about the user's deployment preferences? "
    "Use the memory recall function to check.",
    settings={"function_choice_behavior": "auto"}
)
print(result)

With function_choice_behavior set to "auto", the agent automatically calls memory.recall when it determines context would help, and memory.remember when it encounters facts worth persisting.

Step 4: Direct API Access

from remlabs import RemMemory

mem = RemMemory(api_key="sk-slop-...")

# Store memories outside the kernel
mem.store(
    value="User prefers Azure over AWS for new deployments.",
    namespace="semantic-kernel",
    tags=["preferences", "infrastructure"]
)

# Search with tag filtering
results = mem.search(
    "cloud provider preference",
    namespace="semantic-kernel",
    limit=5
)
for r in results:
    print(r["value"], r["score"])

Triple-Indexed Retrieval

Every memory stored through REM is automatically indexed three ways: vector embeddings for semantic similarity, full-text indexing for exact keyword matching, and entity graph extraction for structured relationships. Multi-signal fusion retrieval combines all three at query time, reaching 90% on LongMemEval -- far beyond what vector-only connectors achieve.

Works with C# too: The REM Labs API is a standard REST API. Use HttpClient to call POST /v1/memory/store and POST /v1/memory/search from any .NET application. See the API docs for the full endpoint reference.

Give your Semantic Kernel agents a memory

Free tier. No credit card. pip install and go.

Get started free →

Memory Plugin for Microsoft Semantic Kernel

Why Semantic Kernel Needs External Memory

Step 1: Install

Step 2: Build the Memory Plugin

Step 3: Register and Use

Step 4: Direct API Access

Triple-Indexed Retrieval

Related articles

Give your Semantic Kernel agents a memory