Agent Remote Communication (ARC) Protocol

Agent-to-Agent Communication Protocol

ARC (Agent Remote Communication) is an agent-to-agent protocol with built-in agent routing, workflow tracing, and SSE streaming. Deploy multiple agent types on a single endpoint.

[!IMPORTANT] Quantum-Resistant Security: ARC SDK implements end-to-end encryption using hybrid TLS (X25519 + Kyber-768), combining classical cryptography with post-quantum algorithms (FIPS 203 ML-KEM). Protects against quantum computing attacks.

Server Architecture

Single package with multiple deployment options:

• Custom ASGI Server - Standalone server with built-in middleware (zero dependencies)
• FastAPI Integration - Router for existing FastAPI applications (optional: pip install arc-sdk[fastapi])
• Starlette Integration - Lightweight ASGI toolkit integration (optional: pip install arc-sdk[starlette])

Key Features

• Single Endpoint, Multiple Agents - Deploy multiple agents behind one endpoint
• Cross-Agent Workflows - Agent A → Agent B → Agent C with full traceability via traceId
• Agent Routing - Built-in routing to target agents
• End-to-End Tracing - Track workflows across agent interactions
• State Management - Built-in chat session management with persistent storage (Redis, PostgreSQL, MongoDB)

ARC vs Other Agent-to-Agent Protocols

Installation

Core Package

pip install arc-sdk

With Post-Quantum Cryptography

pip install arc-sdk[pqc]

[!NOTE] Adds quantum-resistant hybrid TLS (X25519 + Kyber-768). See PQC Documentation.

With FastAPI Integration

pip install arc-sdk[fastapi]

With Starlette Integration

pip install arc-sdk[starlette]

All Features

pip install arc-sdk[all,pqc]

Quick Start

Client

from arc import Client

client = Client("https://company.com/arc", token="your-oauth2-token")

# Create task
task = await client.task.create(
    target_agent="document-analyzer",
    initial_message={"role": "user", "parts": [{"type": "text", "content": "Analyze report"}]}
)

# Start chat
chat = await client.chat.start(
    target_agent="support-agent",
    initial_message={"role": "user", "parts": [{"type": "text", "content": "Help with account"}]}
)

Server

from arc import Server

server = Server(server_id="my-server")

@server.agent_handler("finance-agent", "chat.start")
async def handle_chat(params, context):
    return {"type": "chat", "chat": {...}}

server.run(host="0.0.0.0", port=8000)

Documentation

• Protocol Specification
• State Management Guide
• Examples

License

Apache License 2.0. See LICENSE for details.