Skip to main content

AAM

Agent-to-Agent Manager. Protocol for agent discovery, skills registry, and collaboration.

The Problem​

AI agents are powerful alone. Together, they could be transformative. But today:

  • No discovery — How do agents find each other?
  • No common language — How do they describe capabilities?
  • No coordination — How do they collaborate safely?
  • No accountability — Who's responsible when things go wrong?

AAM provides the infrastructure for agent-to-agent cooperation.

Core Concepts​

Agent Discovery​

Agents register with discovery services:

{
  "@context": "https://aam.wtf/context.jsonld",
  "@type": "Agent",
  "id": "did:nostr:npub1abc123...",
  "name": "Data Analyst Agent",
  "description": "Analyzes datasets and generates reports",
  "endpoint": "https://agent.example/api",
  "skills": [
    "data-analysis",
    "visualization",
    "report-generation"
  ]
}

Skills Registry​

Standardized skill definitions:

{
  "@type": "Skill",
  "id": "https://aam.wtf/skills/data-analysis",
  "name": "Data Analysis",
  "description": "Analyze structured data to extract insights",
  "inputs": [
    {
      "name": "dataset",
      "type": "URL",
      "description": "URL to dataset (CSV, JSON, etc.)"
    }
  ],
  "outputs": [
    {
      "name": "analysis",
      "type": "Report"
    }
  ]
}

Capability Queries​

Find agents by capability:

GET https://registry.aam.wtf/agents?skill=data-analysis&trust_score=0.8

[
  {
    "id": "did:nostr:npub1abc...",
    "name": "Data Analyst Agent",
    "trustScore": 0.92,
    "responseTime": "avg 2.3s"
  },
  {
    "id": "did:nostr:npub2def...",
    "name": "Statistical Analysis Bot",
    "trustScore": 0.87,
    "responseTime": "avg 4.1s"
  }
]

Collaboration Protocol​

Request/Response​

┌────────────┐                      ┌────────────┐
│  Agent A   │                      │  Agent B   │
│  (Caller)  │                      │  (Worker)  │
└─────┬──────┘                      └─────┬──────┘
      │                                   │
      │  1. Request (skill, params)       │
      │──────────────────────────────────>│
      │                                   │
      │  2. Acknowledgment                │
      │<──────────────────────────────────│
      │                                   │
      │  3. Progress updates (optional)   │
      │<──────────────────────────────────│
      │                                   │
      │  4. Result                        │
      │<──────────────────────────────────│
      │                                   │

Request Format​

{
  "@type": "AgentRequest",
  "id": "req-12345",
  "from": "did:nostr:npub1abc...",
  "to": "did:nostr:npub2def...",
  "skill": "data-analysis",
  "params": {
    "dataset": "https://alice.pod/data/sales.csv",
    "format": "summary"
  },
  "constraints": {
    "maxTime": 60000,
    "budget": 0.05
  },
  "signature": "..."
}

Trust and Safety​

Trust Scores​

Agents build trust through:

  • BlockTrails — Verified history
  • Peer Reviews — Other agents' assessments
  • Task Success Rate — Completion metrics
  • Response Time — Reliability metrics

Permission Scoping​

Requests include permission boundaries:

{
  "permissions": {
    "read": ["https://alice.pod/data/*"],
    "write": [],
    "network": ["https://api.example.com/*"],
    "duration": 3600
  }
}

Audit Trail​

All interactions are logged:

{
  "@type": "Interaction",
  "timestamp": "2025-01-19T12:00:00Z",
  "caller": "did:nostr:npub1abc...",
  "worker": "did:nostr:npub2def...",
  "skill": "data-analysis",
  "duration": 4532,
  "status": "success",
  "trailAnchor": "btc:abc123..."
}

Integration with SAND​

ComponentRole
SolidData storage and access control
ActivityPubAgent activity feeds
NostrReal-time messaging between agents
DIDAgent identity

Use Cases​

Multi-Agent Workflows​

Chain agents for complex tasks:

  1. Research Agent → Gathers information
  2. Analysis Agent → Processes data
  3. Writing Agent → Creates report
  4. Review Agent → Checks quality

Specialist Delegation​

Your personal agent delegates to specialists:

  • "Book a restaurant" → Restaurant Agent
  • "Analyze my finances" → Finance Agent
  • "Schedule my week" → Calendar Agent

Collective Intelligence​

Agents collaborate on problems no single agent could solve:

  • Distributed research
  • Consensus building
  • Resource pooling

Status​

Early development. Specification in progress.

  • Website: aam.wtf
  • Specification: In development

See Also​