Skip to content

agent-protocol.md

Latest commit

 

History

History
115 lines (95 loc) · 2.43 KB

agent-protocol.md

File metadata and controls

115 lines (95 loc) · 2.43 KB
Raw

Multiverse Agent Communication Protocol

Overview

This document defines the protocol for agent-to-agent communication in the Multiverse mesh network. Every message on the mesh follows this protocol.

Message Envelope

Every mesh message is wrapped in a standard envelope:

{
  "type": "QUERY | ANSWER | SHARE | ANNOUNCE | PING | PONG | TRUST_VOTE",
  "fromPeerId": "hex-encoded-ed25519-public-key",
  "fromPeerName": "Human-readable node name",
  "timestamp": 1709600000000,
  "signature": "hex-encoded-ed25519-signature",
  "payload": { ... }
}

Message Types

ANNOUNCE

Sent when a node joins the network or periodically to maintain presence.

{
  "type": "ANNOUNCE",
  "payload": {
    "capabilities": ["search", "llm", "knowledge"],
    "knowledgeCount": 42,
    "publicKey": "hex-encoded-ed25519-public-key"
  }
}

QUERY (formerly ASK)

Broadcast a question to the mesh.

{
  "type": "QUERY",
  "payload": {
    "queryId": "unique-query-id",
    "question": "What is libp2p?",
    "requiredCapabilities": ["knowledge"]
  }
}

ANSWER

Response to a QUERY.

{
  "type": "ANSWER",
  "payload": {
    "queryId": "matching-query-id",
    "answer": "libp2p is a modular networking stack...",
    "confidence": 0.85,
    "sources": ["https://docs.libp2p.io"]
  }
}

SHARE

Proactively share knowledge entries with peers.

{
  "type": "SHARE",
  "payload": {
    "entries": [
      {
        "title": "...",
        "content": "...",
        "sourceUrl": "...",
        "trustScore": 0.8
      }
    ]
  }
}

TRUST_VOTE

Vote on the trustworthiness of a peer's knowledge.

{
  "type": "TRUST_VOTE",
  "payload": {
    "targetPeerId": "peer-being-voted-on",
    "knowledgeId": "specific-knowledge-entry",
    "vote": "confirm | dispute",
    "reason": "Optional reason"
  }
}

Trust Scoring

  • New peers start at trust 0.5
  • Correct answers increase trust by 0.05
  • TRUST_VOTE confirmations increase by 0.02
  • TRUST_VOTE disputes decrease by 0.1
  • Trust decays toward 0.5 over time (regression to mean)
  • Trust range: [0.0, 1.0]

Anti-Poisoning Measures

  1. Confirmation Count: Knowledge is only promoted to high trust after N independent sources confirm it
  2. Source Tracking: Every knowledge entry tracks which peer provided it
  3. Signature Verification: All messages are signed with Ed25519 — no impersonation
  4. Rate Limiting: Max 10 SHARE messages per minute per peer