MemOS Tutorial: Deploy Self-Evolving Memory for Your AI Agents

MemOS Tutorial: Deploying a Self-Evolving Memory System for Your AI Agents

In this tutorial, we'll walk through deploying MemOS — a production-ready Memory Operating System — and integrating it with your AI agents. By the end, you'll have a fully functional long-term memory backend that learns and evolves with your agents.

🚀 Want to deploy MemOS yourself?

Docker configs, system requirements, and installation guides — all on one page.

View MemOS Tool Page →

Prerequisites

A server with at least 4GB RAM and 2 CPU cores (8GB/4CPU recommended)
Docker and Docker Compose installed
Python 3.10+ (for the local plugin)
An LLM API key (OpenAI, Anthropic, or local model)

Step 1: Quick Start with Docker

The fastest way to get MemOS running is via Docker Compose. Create a docker-compose.yml file:

services:

  memos:

    image: ghcr.io/memtensor/memos:latest

    restart: unless-stopped

    ports:

      - 8080:8080

    volumes:

      - ./data/memos:/data

Then run:

docker compose up -d

MemOS will be available at http://localhost:8080.

Step 2: Configure Memory API

MemOS provides a RESTful Memory API. Here's how to store your first memory:

curl -X POST http://localhost:8080/api/v1/memory \

  -H "Content-Type: application/json" \

  -d '{

    "content": "User prefers concise technical answers",

    "type": "preference",

    "metadata": {"source": "conversation", "confidence": 0.9}

  }'

Step 3: Multi-Agent Memory Sharing

MemOS excels at multi-agent scenarios. Different agents can share and contribute to the same memory store while maintaining isolation through memory cubes:

# Agent A stores a tool-specific memory

curl -X POST http://localhost:8080/api/v1/memory \

  -d '{"content":"Docker restart needed after config change","cube":"devops","agent":"agent-a"}'

# Agent B retrieves from the same cube

curl -X GET "http://localhost:8080/api/v1/memory?cube=devops&query=docker+restart"

Step 4: Integrate with Hermes Agent

MemOS has an official local plugin for Hermes Agent. Install it and enable hybrid retrieval (FTS5 + vector) with smart deduplication:

pip install memos-local-plugin

# Configure in your Hermes Agent profile

memos-plugin --enable --retrieval hybrid --dedup smart

💡 Tip: MemOS uses tiered skill evolution — the more your agents use it, the more it learns. Start with L1 traces and watch crystallized skills emerge over time.

Real-World Impact

In production benchmarks, MemOS improved OpenClaw task completion rates from 36.63% to 50.87% across five agent tasks — a 39% relative improvement. With 35.24% token savings through smart deduplication and hybrid retrieval, it pays for itself in API cost reductions alone.

🚀 Ready to deploy MemOS?

Get the Docker configs, system requirements, and full installation guide.

View MemOS Tool Page →