Skip to content
Rajamohan J edited this page Mar 5, 2026 · 2 revisions

Agent Transport Protocol (ATP) — Wiki

The TCP/IP of AI Agents

Welcome to the ATP Wiki — the comprehensive guide to understanding, using, and contributing to the Agent Transport Protocol.


What is ATP?

ATP is a five-layer protocol stack for trust-aware, economically-optimal multi-agent networking. It provides the missing networking layer that allows AI agents to:

  • Identify each other with cryptographic DIDs
  • Trust each other with time-decayed scoring
  • Negotiate capabilities with binding QoS contracts
  • Compress context by 28x using semantic differentials
  • Route tasks optimally across agent economies
  • Recover from failures with circuit breakers and heartbeats

Quick Navigation

Page Description
Getting Started Install, build, and run your first ATP program
Architecture Overview The five-layer protocol stack explained
Layer 1: Identity and Trust Ed25519 DID, trust scoring, Sybil resistance
Layer 2: Capability Handshake 3-phase negotiation and QoS contracts
Layer 3: Context Compression Semantic Context Differentials (SCD)
Layer 4: Economic Routing Bellman-Ford, 5 routing patterns
Layer 5: Fault Tolerance Circuit breaker, heartbeat, checkpoint
SDK API Reference Complete function reference for atp-sdk
Python SDK PyO3 bindings and Python usage
Benchmarks AgentNet-Bench results and analysis
gRPC Service Protobuf definitions and wire protocol
ATP vs MCP vs RAG Why ATP exists and how it compares
Case Studies Real-world use cases
Contributing How to contribute to ATP
FAQ Frequently asked questions

Key Numbers

Cost Reduction:     -53.4% vs sequential
Context Compression: 28x via SCD
Task Failures:       0 / 10,000
Quality Score:       0.904 (+8% over baselines)
Latency Reduction:  -29.3%
Routing Decisions:  < 1 microsecond
Tests:               280 passing, zero failures
Lines of Rust:      ~37,000 across 75 files

Links


Created by Rajamohan Jabbala — AlphaForge AI Labs

Clone this wiki locally