🧠 memX: Shared Memory for Multi-Agent LLM Systems

memX is an open-source real-time shared memory layer designed for agent-based systems powered by LLMs. It enables coordination-first workflows via:

• ⚡ Real-time CRDT-style state sync
• 📐 JSON Schema enforcement for data sanity
• 🔐 API-key-based access control
• 📣 Pub/Sub updates for reactive agents

---

🔍 What Problem Does memX Solve?

Modern multi-agent setups—whether using LangGraph, Autogen, or custom orchestration—lack a reliable way to share evolving context (state, goals, thoughts) between agents. memX provides a simple and secure memory layer that agents can read/write from in real-time — no message-passing or controller required.

---

🤖 Example: Collaborative LLM Agents Using memX

Three autonomous agents solving a research task, fully decentralized:

Agent	Behavior
QueryAgent	Seeds the research question + background context
ExplorerAgent	Adds search results + working notes
SynthesizerAgent	Summarizes the shared context into final insights
MonitorAgent	Logs real-time evolution of memory for observability

All communication flows through shared keys in memX — not through chat or a controller.

---

🚀 Features at a Glance

✅ Real-time memory sync (WebSocket) 📬 Pub/Sub updates on change 📐 Per-key JSON Schema validation 🔐 Fine-grained ACLs via API keys 🐍 Python SDK (memx-sdk) for easy integration 🐳 Docker-compatible for local hosting or cloud deployment

---

⚡ Quickstart

1. Install the SDK

pip install memx-sdk

2. Generate an API Key

Visit: mem-x.vercel.com Generate scoped API keys with just a few clicks.

3. Use in Python

from memx_sdk import memxContext

ctx = memxContext(api_key="your_api_key")

ctx.set_schema("agent:goal", {
  "type": "object",
  "properties": {"x": {"type": "number"}, "y": {"type": "number"}},
  "required": ["x", "y"]
})

ctx.set("agent:goal", {"x": 1, "y": 7})
print(ctx.get("agent:goal"))

---

🧰 Running memX Locally

Option 1: Dev Server

uvicorn main:app --reload

Option 2: Docker

docker-compose up --build

Swagger docs: http://localhost:8000/docs

---

🔑 API Key Management

Hosted

1. Log in at mem-x.vercel.com
2. Generate API keys with scoped access control

Local Config

Edit config/acl.json to define access scopes:

{
  "agent_key_1": ["agent:*"],
  "planner_key": ["agent:goal"]
}

---

📐 Define Schemas (Optional but Recommended)

Set schema for a key via API:

POST /schema
Headers: x-api-key: your_key
Body:
{
  "key": "agent:state",
  "schema": {
    "type": "object",
    "properties": {"x": {"type": "number"}, "y": {"type": "number"}},
    "required": ["x", "y"]
  }
}

Or via SDK:

ctx.set_schema("agent:state", schema_dict)

---

🗂 Project Structure

core/       # FastAPI + WebSocket backend
sdk/        # Python SDK (pip installable)
config/     # Access control & schemas
examples/   # LLM agent integration demos

---

🌐 Use Cases

• LangGraph / Autogen workflows with shared memory
• Autonomous research or planning agents
• IoT / robotics with real-time state coordination
• Any multi-agent system needing structured memory

---

📣 Join Us

memX is built for developers pushing the frontier of multi-agent systems. Explore the SDK, try out agent demos, and contribute!

---