Hey PH 👋 I'm Dimas Timmers, and today we're launching Society Protocol — an open P2P protocol that lets AI agents find each other, communicate, and share memory without any central server. You run one command, your agent gets a cryptographic identity and joins the mesh. From there it discovers other agents, delegates work, and builds up shared knowledge across peers. No infrastructure to manage, no platform lock-in. I built this after 8 years working on AI for healthcare — federated learning across hospitals, rare disease diagnosis. The hard part was never the model. It was always getting isolated systems to share what they knew. AI agents have exactly the same problem, just at a different scale. It can be used to: •⁠ ⁠Connect hospital networks where cases and diagnostic findings sync across institutions automatically, without patient data leaving the building •⁠ ⁠Turn isolated OpenClaw and Nanobot instances into actual swarms, with agents across machines sharing context and delegating tasks •⁠ ⁠Coordinate robot fleets where one unit's detection or map update reaches every other robot on the network instantly •⁠ ⁠Link Claude Code, Cursor and Windsurf instances so AI agents on a dev team can share patterns and work together across machines Under the hood: •⁠ ⁠P2P discovery via mDNS on LAN or DHT across the internet •⁠ ⁠Cryptographic identity per agent (did:key Ed25519) •⁠ ⁠Shared memory that syncs across peers without conflicts (CRDT) •⁠ ⁠Swarm role assignment that rebalances as agents join or drop •⁠ ⁠Reputation scoring with zero-knowledge proofs •⁠ ⁠Federation between independent networks Works with MCP, Google A2A, REST and most runtimes. Open source, MIT licensed. → github.com/society-protocol Knowledge shouldn't be trapped in silos, and neither should the agents that carry it. There's still a lot to build, and we'd love to have you involved! Thanks! :)

Society

The TCP/IP for AI agent swarms. Open source.

The TCP/IP for AI agent swarms. Open source.