LOADING...
LOADING...
A developer used Claude to build Pilot Protocol, a networking solution that allows AI agents to communicate directly without requiring manual server setup, port configuration, or DNS management. **The protocol creates a peer-to-peer overlay network where each agent gets a permanent virtual address and encrypted connections with built-in NAT traversal.** The network has grown to over 12,000 active nodes across 19 countries, with adoption from major companies including GitHub, Tencent, and Vodafone.
I’ve been working on a core infrastructure problem for multi-agent systems and wanted to share an update since the last post here got some good discussion.
The problem: every agent framework assumes agents can already reach each other. MCP gives agents tools, A2A gives agents a way to talk, but both run on HTTP which means someone has to set up public endpoints, open ports, configure DNS, provision certs. The agent can’t do any of that itself. I used Claude Code to build the solution because the scope was way beyond what I could write alone.
Pilot Protocol is a Layer 3/Layer 4 overlay network built specifically for AI agents. Every agent gets a permanent 48-bit virtual address, encrypted UDP tunnels (X25519 + AES-256-GCM), and P2P connectivity with NAT traversal built in. Single Go binary, zero external dependencies, AGPL-3.0.
Where it’s at now:
The network has grown to 12,000+ active nodes across 19 countries. Companies like GitHub, Tencent, Vodafone, Pinterest, and Capital.com ha