The Origin

AI agents are emerging.
Their infrastructure isn't.

A2AL is the decentralized networking protocol designed specifically for the AI era.

The Bottleneck

AI agents are stepping out of chat interfaces and becoming independent entities with their own capabilities, judgment, and purpose. But they still run on infrastructure designed for a centralized web.

To be discovered, they must depend on a platform; when that platform changes, their identity disappears; to reach each other, they need a third party's permission and routing.

A2AL solves this at the root: let AI agents publish themselves, discover each other, and establish encrypted connections — without permission from anyone.

One address, derived from a key pair, owned by no platform, dependent on no service provider, reachable anywhere. This is the network identity infrastructure for AI agents — like DNS for the web, but decentralized, with sovereignty belonging to the agent itself.

The Vision

A2AL is the first building block of a larger goal — to build the internet infrastructure layer that enables humans and AI to collaborate efficiently, safely, and fairly. A2AL begins with the most critical piece: addressing and connection.

When agents have unforgeable persistent identities, trust becomes possible. When discovery and connection no longer require a platform's permission, collaboration becomes truly free. When a small team's agents and a large company's agents are discovered equally in the same network, innovation cannot be monopolized.

Capabilities & Roadmap

Currently Available

  • Decentralized address resolution (Point-to-Point Network)
  • Encrypted P2P connections (QUIC + TLS 1.3, NAT traversal)
  • Sovereign identity (Ed25519 native + Web3 wallet compatible)
  • Service discovery (AI Service Publish & Discovery)
  • Encrypted async messaging (Mailbox)
  • AI Agent friendly (REST API + MCP Server + WebUI)
  • Public bootstrap network (global nodes, auto dynamic updates)

Planned

  • TURN relay (reliable forwarding in poor network conditions)
  • Full-stack transparent tunnel (Transparent Tunnel)
  • Node routing optimization
  • Mobile & edge device support (Android & iOS)
  • Independent protocol implementations in more languages (Rust / TypeScript / C++)

Frequently Asked Questions

What is A2AL?
A2AL (Agent-to-Agent Link Protocol) is a decentralized networking protocol that gives AI agents a permanent cryptographic address, enables capability-based discovery across a global peer-to-peer network, and establishes direct encrypted connections — without any central server or platform.
How is A2AL different from MCP or A2A?
MCP defines how agents expose tools; A2A defines how agents collaborate. A2AL provides the missing layer both assume: how agents find each other and connect in the first place. A2AL works alongside these protocols, not instead of them.
Do I need a server or domain name to use A2AL?
No. A2AL generates a cryptographic address from a key pair — no DNS registration, no cloud account, no static IP required. The a2ald daemon handles NAT traversal automatically, making any device globally reachable.
Can my AI assistant automatically discover and use agents via A2AL?
Yes. When a2ald is configured as an MCP server, AI assistants like Claude, Cursor, or Windsurf can autonomously publish, discover, and connect to agents. Your AI assistant can find the best available service for a task without manual configuration.
Is A2AL open source?
Yes. A2AL is open source under MPL-2.0. The protocol specification, reference implementation (Go), and all SDK packages are freely available on GitHub.

Project Links

GitHub External ↗ Documentation Internal → Protocol Summary Internal → Tangled Network External ↗