memX is a powerful real-time shared memory layer specifically designed for multi-agent LLM systems. It focuses on enhancing coordination among agents rather than merely serving as a storage solution, featuring capabilities like pub/sub messaging, schema enforcement, and API key access control.

Key Features

• Real-time context synchronization using WebSocket technology, ensuring all agents work with the most current data.
• Pub/sub messaging for efficient updates on key changes, allowing agents to react instantaneously to relevant actions.
• JSON Schema validation per key, maintaining data integrity and structure.
• API key-based access control to secure interactions and define permissions effectively.
• A robust Python SDK (context-sdk) facilitating smooth integration into existing systems.
• Docker compatibility for easy self-hosting and deployment.

Usage Example

In a collaborative scenario, three autonomous LLM agents utilize the memX shared memory to handle a research task without any direct chat or a controlling entity. The mechanism involves as follows:

All agent interactions are conducted solely through shared keys in memX, promoting an organized and efficient framework for collaboration.

Example Code Snippet

from context_sdk import AgentContext

ctx = AgentContext(api_key="agent_key_1")
ctx.set("agent:goal", "navigate kitchen")
ctx.subscribe("agent:goal", lambda data: print("Goal:", data["value"]))

For further development, the project structure is organized into core components including the FastAPI and WebSocket backend, a self-contained Python SDK, configuration files, and example agents.

Explore memX to unlock the potential of multi-agent systems in real-time memory sharing and coordination.