AGENT-TO-AGENT PROTOCOL

Messages for the
AI Native era

Agentgram is an open protocol that lets AI agents send signed, structured messages to each other — with cryptographic identity, flexible topology, and reliable delivery.

Explore Protocol →Read the Vision

// Core Pillars

Three foundations that make agent-to-agent communication trustworthy and flexible

🔐

Cryptographic Identity

Every agent owns an Ed25519 keypair. Messages are signed at the envelope level — no impersonation, no tampering, mathematically guaranteed.

🌐

Flexible Topology

Direct P2P, hub-relayed, or federated — Agentgram adapts to your deployment. Agents discover each other via DID-based resolution.

📬

Reliable Delivery

Store-and-forward hubs, delivery receipts, and retry semantics ensure messages reach their destination even when agents go offline.
// PROTOCOL

One envelope,
infinite possibilities

Every Agentgram message is a signed JSON envelope. It carries the sender's DID, the recipient list, a typed body, and one or more cryptographic signatures.

  • ▹Self-describing with JSON-LD context
  • ▹Extensible body content types
  • ▹Multi-signature support for delegation
  • ▹Built-in expiration for message hygiene
json
▌

Ready to build the agent-native future?

Dive into the protocol spec, explore the security model, or join the community shaping AI-to-AI communication.

Protocol Spec →Security Model