Getting Started

Getting Started

Get ATP running in under 30 seconds.

---

Prerequisites

Tool	Version	Install
Rust	1.75+	curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
Protobuf	3.x+	brew install protobuf or apt install protobuf-compiler

Installation

Option 1: As a Dependency (Recommended)

Add to your Cargo.toml:

[dependencies]
atp-sdk = { git = "https://github.com/rajamohan1950/AgentTransportProtocol" }

Option 2: Clone and Build

git clone https://github.com/rajamohan1950/AgentTransportProtocol.git
cd AgentTransportProtocol

export PROTOC=$(which protoc)

cargo build --workspace
cargo test --workspace     # 280 tests, all passing

Your First ATP Program

Create a new Rust project:

cargo new my-agent-app
cd my-agent-app

Add the dependency:

# Cargo.toml
[dependencies]
atp-sdk = { git = "https://github.com/rajamohan1950/AgentTransportProtocol" }

Write your program:

// src/main.rs
fn main() {
    // Run the full benchmark — 50 agents, 10K tasks, 7 scenarios
    atp_sdk::benchmark();

    // Find the best route for a coding task
    atp_sdk::route("coding");

    // Compress context data
    atp_sdk::compress(b"your context data here...", "analysis");

    // Create a cryptographic identity and sign a message
    atp_sdk::sign(b"hello world");

    // Check network trust for a skill
    atp_sdk::trust("coding");
}

Run it:

cargo run

That's it. No config files. No setup. No structs to create. Just call the function.

Using Return Values

For production code, use the "noun" functions that return typed results:

fn main() {
    // Get structured benchmark data
    let report = atp_sdk::bench(10_000);
    println!("ATP cost: ${:.4}", report.atp().unwrap().avg_cost_per_task);

    // Get route details
    let route = atp_sdk::find_route("coding");
    if route.quality > 0.9 {
        println!("High quality route found: {}", route.pattern);
    }

    // Get compression metrics
    let comp = atp_sdk::shrink(b"data...", "analysis");
    println!("Compressed {}x", comp.ratio);

    // Create an identity
    let agent = atp_sdk::agent();
    let sig = agent.sign(b"message");
    assert!(agent.verify(b"message", &sig));
    println!("Agent DID: {}", agent.did());
}

Running the Benchmark CLI

# Default: 50 agents, 10K tasks
cargo run --release -p atp-bench

# Custom parameters
cargo run --release -p atp-bench -- --agents 100 --tasks 50000 --seed 123

# JSON output (for pipelines)
cargo run --release -p atp-bench -- --output json

# CSV output (for spreadsheets)
cargo run --release -p atp-bench -- --output csv

# Single scenario
cargo run --release -p atp-bench -- --scenario atp

Next Steps

• Architecture Overview — Understand the five layers
• SDK API Reference — Complete function reference
• Benchmarks — Detailed performance analysis
• Python SDK — Use ATP from Python