ARP - Agent Resource Protocol

Unified resource protocol for AI Agents ｜ AI Agent 统一资源协议

---

Overview

ARP (Agent Resource Protocol) is a URL scheme designed specifically for AI Agents to reference and access resources in a unified way. It separates what a resource is from how to get it.

URL Format

arp:{semantic}:{transport}://{location}

Components

Component	Description	Examples
semantic	What the resource is	text, json, image, prompt
transport	How to fetch it	https, http, file, s3
location	Where to find it	example.com/file.txt, /local/path

Examples

arp:text:https://example.com/readme.txt
arp:json:file:///config/settings.json
arp:image:s3://bucket/avatar.png
arp:prompt:https://prompts.example.com/assistant.md

Design Principles

1. Separation of Concerns

• Semantic layer: Defines how to interpret the content
• Transport layer: Defines how to fetch the content
• Location: Defines where the content is

2. Extensibility

Both semantic and transport layers are extensible:

arp:{custom-semantic}:{custom-transport}://{location}

3. Agent-First

Designed for AI Agents to:

• Declare resource dependencies explicitly
• Abstract away infrastructure details
• Enable unified resource management

Semantic Types

Core Types

Type	Description	Content
text	Plain text	String
json	JSON data	Parsed object
image	Image file	Binary/Base64
prompt	Prompt template	String with variables

Extended Types (Future)

Type	Description
audio	Audio content
video	Video content
code	Source code with syntax
embedding	Vector embedding

Transport Types

Core Types

Type	Description
https	HTTPS protocol
http	HTTP protocol
file	Local filesystem

Extended Types (Future)

Type	Description
s3	AWS S3
gcs	Google Cloud Storage
ipfs	IPFS protocol
arr	Agent Resource Registry

Resolution Flow

┌─────────────────────────────────────────────────────────┐
│                    ARP URL                              │
│         arp:text:https://example.com/file.txt           │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                      Parse                              │
│    semantic: text, transport: https, location: ...      │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                Transport Layer                          │
│              fetch(location) → Buffer                   │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                 Semantic Layer                          │
│            parse(buffer) → Resource                     │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                    Resource                             │
│          { type, content, meta }                        │
└─────────────────────────────────────────────────────────┘

Implementation

The reference implementation is ResourceX.

npm install resourcexjs

import { createResourceX } from "resourcexjs";

const rx = createResourceX();
const resource = await rx.resolve("arp:text:https://example.com/file.txt");

Versioning

This specification follows Semantic Versioning.

Current Version: 0.1.0

Ecosystem

ARP is part of the Deepractice AI Agent infrastructure:

• PromptX - AI Agent context platform
• AgentX - AI Agent runtime platform
• ResourceX - Unified resource manager (ARP)
• ToolX - Tool integration
• UIX - AI-to-UI protocol layer
• SandboX - Agent sandbox

License

MIT