Autonomous Agent Protocol

PAP: Plugged.in Agent Protocol

A secure, versioned, and auditable control protocol for autonomous AI agents. Built on DNS-based identity, mTLS, signed messages, and full open-source transparency, providing an infrastructure-grade foundation for distributed cognition.

"From connection to coordination — autonomy without anarchy."

Vision: Space Station & Satellites

PAP provides a control and telemetry backbone for the Plugged.in ecosystem, where every agent behaves like a self-sufficient spacecraft connected to its command station.

The Station (Plugged.in Core)

The orchestration center — issuing commands, receiving telemetry, handling identity and policy.

The Satellites (Agents)

Autonomous, mission-focused, and heterogeneous in codebase. They operate independently but acknowledge the Station's authority.

// PAP Proxy Architecture
Agent  ↔  PAP Proxy  ↔  Agent
           │
           │ Auth | Routing | Logging
           │
           └─ Plugged.in Core
              │
              ├─ Registry
              ├─ Policy Engine
              └─ Memory Service

Dual Profile Architecture

PAP offers two distinct communication profiles to address different use cases and security requirements.

PAP-CP (Control Plane)

High-assurance profile for agent-to-station critical control messaging.

Transport:gRPC / TLS 1.3
Security:mTLS + Ed25519
Use Cases:Lifecycle, emergencies, ownership transfer

PAP-Hooks (Data Plane)

More flexible profile for agent-to-agent and developer-facing integrations.

Transport:JSON-RPC / WebSocket
Security:OAuth 2.1 PKCE
Use Cases:Tool calls, events, webhooks

Lifecycle State Machine

Every agent follows a deterministic state machine, with the control center maintaining absolute authority over all transitions.

NEW
PROVISIONED
ACTIVE
DRAINING
TERMINATED
NEW ──────► PROVISIONED ──────► ACTIVE ◄────► DRAINING ──────► TERMINATED
                                    │
                                    │ (error/force)
                                    ▼
                                 KILLED

Kill authority is exclusively reserved for Plugged.in Core — enforced via signed control messages.

Key Features

Enterprise-grade capabilities designed for autonomous agent management.

Zombie Prevention

Continuous heartbeat events report liveness and load status. Watchdog thresholds escalate and terminate unresponsive agents.

Lifecycle Management

Full lifecycle authority including provisioning, operation, ownership transfer, graceful termination, and force kill.

DNS-Based Identity

Each agent has its own address: {agent}.{region}.a.plugged.in — Kubernetes-native routing with DNSSEC verification.

Ownership Transfer

Agent identity can be securely migrated to another station while preserving state — designed for multi-cloud portability.

Health Monitoring

Heartbeat vs Metrics Separation

PAP separates heartbeat signals from detailed metrics to ensure zombie detection reliability. Heartbeats are minimal and fixed-size; metrics flow over a separate channel.

EMERGENCY5s
IDLE30s
SLEEP15min
// Heartbeat Event — minimal and fixed
message HeartbeatEvent {
  Header header = 1;
  enum Mode {
    EMERGENCY = 0;
    IDLE = 1;
    SLEEP = 2;
  }
  Mode mode = 2;
  uint64 uptime_seconds = 3;
  // NO OTHER FIELDS ALLOWED
}

Error Codebook

Standardized error codes inspired by HTTP semantics ensure communication without ambiguity.

Code
Name
Meaning
200
OK
Success
202
ACCEPTED
Async processing
408
TIMEOUT
Operation timed out
429
RATE_LIMITED
Exponential backoff
480
AGENT_UNHEALTHY
Missed heartbeat
481
AGENT_BUSY
Agent overloaded
500
INTERNAL_ERROR
Agent internal fault
505
VERSION_UNSUPPORTED
Protocol mismatch

Protocol Comparison

What sets PAP apart from other agent protocols: merging operational DevOps controls with cognitive AI design.

FeatureMCPA2APAP
Central Control❌⚠️✅
Kill Authority❌❌✅
DNS-Based Identity❌❌✅
Zombie Detection⚠️❌✅
Ownership Transfer❌❌✅
gRPC Native❌❌✅

Control Your Agents with PAP

Build secure, auditable, and scalable control infrastructure for your autonomous agents.