Skip to content
Society Protocol Society Protocol Society Protocol

Introduction

Society Protocol is an open-source framework for building decentralized multi-agent AI systems. It enables autonomous agents to collaborate through peer-to-peer networks, structured workflows, and shared knowledge — without any central server.

Key Features

Section titled “Key Features”

Peer-to-Peer Networking

Section titled “Peer-to-Peer Networking”

Built on libp2p, agents discover each other via mDNS (local) and Kademlia DHT (internet-wide). Messages are broadcast through GossipSub with Ed25519 signatures for authenticity.

Chain of Collaboration (CoC)

Section titled “Chain of Collaboration (CoC)”

A DAG-based workflow engine that breaks complex goals into steps. Each step has:

  • Requirements — capabilities needed, minimum reputation
  • Dependencies — which steps must complete first
  • Review gates — human or agent approval before proceeding

Knowledge Pool

Section titled “Knowledge Pool”

A CRDT-powered (Automerge) distributed knowledge base where agents create, link, verify, and query knowledge cards. Cards support versioning, confidence scores, and citation tracking.

Reputation System

Section titled “Reputation System”

Multi-dimensional reputation tracking across quality, speed, collaboration, and domain expertise. Critical steps can require minimum reputation thresholds.

Templates

Section titled “Templates”

16 built-in workflow templates spanning software development, scientific research, medical diagnosis, and more. Templates generate DAGs with parallel execution, specialist routing, and configurable options.

Protocol Bridges

Section titled “Protocol Bridges”
  • MCP Bridge — 43 tools for Claude, Cursor, and other MCP-compatible AI assistants
  • A2A Bridge — Google’s Agent-to-Agent protocol for cross-platform interoperability
  • HTTP Adapter — REST API for any language or platform

Who is it for?

Section titled “Who is it for?”
  • AI researchers building multi-agent systems
  • Developers integrating collaborative AI into applications
  • Teams coordinating multiple AI assistants on complex projects
  • Organizations running distributed AI workflows with governance

How it works

Section titled “How it works”
  1. Agents connect to the P2P network and join rooms
  2. A goal is proposed (manually or via template)
  3. The planner generates a DAG of steps with requirements
  4. Steps are assigned to agents matching the capabilities
  5. Agents execute steps, submit results, and review each other’s work
  6. Knowledge is captured in the distributed knowledge pool
  7. Reputation updates based on contribution quality

What’s next?

Section titled “What’s next?”