OPEN STANDARD • MIT LICENSED
85% token reduction. 10x faster discovery. Full observability.
ARW provides efficient, observable infrastructure for AI agents — structured discovery, machine-readable content, and OAuth tools that preserve your SEO and user experience.
BY THE NUMBERS
See how ARW transforms agent-web interactions with concrete efficiency gains
From 55KB HTML to 8KB Markdown
Single manifest vs full site crawl
Reduced bandwidth & infrastructure
IMPLEMENTATION
The web/HTTP layer remains the universal entry point agents use to discover specialized features. ARW provides a standardized way for AI agents to discover content, tools, and integrations.
1# CloudCart Electronics
2
3> E-commerce platform for electronics
4
5We sell computers, peripherals, accessories.
6
7## Products
8
9[Keyboards](/products/keyboards): Browse keyboards
10[Mice](/products/mice): Browse mice
11[Monitors](/products/monitors): Browse monitors
12... (50+ product categories)
13
14## Support
15
16[Shipping Info](/support/shipping): Details
17[Returns](/support/returns): Policy
18... (20+ support pages)FOR EVERYONE
For Publishers
Agents are already browsing your site. ARW ensures they get accurate information, attribute your brand, and can complete transactions.
For AI Platforms
Stop scraping HTML and guessing at structure. ARW provides clean, structured data that reduces errors and lowers costs.
For Builders
Add llms.txt in 2 hours. Progressive enhancement—each step adds value without breaking existing functionality.
PROGRESSIVE ENHANCEMENT
Progressive enhancement means you can start with basic discovery and add features incrementally. Time to first value: 2 hours.
llms.txt advertises ACP checkout endpoints, enabling discovery via the web layer while ACP handles standardized transaction flows.llms.txt and enforced via HTTP headers. Training and inference permissions are specified per-site, with clear attribution requirements and rate limits.PROGRESSIVE CONFORMANCE
Implement incrementally. Each level builds on the previous, adding features without breaking existing functionality.
Discovery Ready
Basic discovery via llms.txt + machine views. Agents can find your content.
2 hours to implement
Semantic Ready
Adds chunking, attribution, and rate limits. Agents can understand your content deeply.
1-2 weeks
Action Ready
OAuth tools + Schema.org semantics. Agents can act safely with consent.
4-6 weeks
Protocol Ready
Full protocol interoperability (MCP, ACP, A2A). Universal agent access.
2-3 months
Zero-config deployment on Cloudflare's global edge network. Drop in a script tag, get instant ARW compliance with automatic content extraction, embedding generation, and semantic search.
Add one script tag. Content automatically extracted and served as .md machine view files.
D1 database, KV cache, Vectorize embeddings, Workers AI—all on 300+ global locations.
Track agent traffic, measure conversions, understand which agents drive value.
┌─────────────────────────────────────────┐ │ ARW Cloud Architecture │ ├─────────────────────────────────────────┤ │ │ │ Website Cloudflare Edge │ │ ┌─────────┐ ┌─────────────────┐ │ │ │arw-loader├────►│ Edge Worker │ │ │ │ .js │POST │ │ │ │ └─────────┘ │ ┌──────┐┌─────┐│ │ │ │ │Workers││ KV ││ │ │ AI Agent │ │ AI ││Cache││ │ │ ┌─────────┐ │ └──┬───┘└──┬──┘│ │ │ │GET .llm.├────►│ │ │ │ │ │ │ .md │ │ ▼ │ │ │ │ └─────────┘ │ ┌──────┐ │ │ │ │ │ │Vector│ │ │ │ │ │ │ize │ │ │ │ │ │ └──────┘ │ │ │ │ └────────────┼────┘ │ │ │ │ │ ◄────────────────────►│ │ │ Sub-50ms latency globally │ └─────────────────────────────────────────┘
USE CASES
From e-commerce to news publishers, ARW provides the infrastructure layer for AI agent interactions across verticals.
Products appear in AI recommendations with attribution. OAuth checkout protects your transaction flow while enabling conversions.
Content helps users via agents while protecting your business model. Excerpts drive traffic back to your site.
Documentation helps developers via agents. Trial signups and support tickets flow through OAuth for tracking.
Patient-facing content with strict training policies. Appointment booking through verified OAuth flows.
Market data and educational content. Strong compliance controls with audit trails for all agent interactions.
API docs and code examples optimized for coding agents. Enable training on open source, protect proprietary content.
TECHNOLOGY
No proprietary protocols. ARW uses technologies that every platform already supports.
Universal protocol layer
Standard HTTP methods and headers. Works with any web framework, CDN, or hosting provider.
Human and machine readable
llms.txt uses YAML for structured data. Machine views use Markdown for efficient content delivery.
Industry-standard auth
Standard OAuth flows for tools. User consent required. Technically enforced permissions.
Semantic vocabulary
JSON-LD with Schema.org types for tools. Compatible with existing SEO and structured data.
Rust-based CLI for generating, validating, and managing ARW files.
validate-arw.pyvalidate-arw.mjsarw_model.yamlARW is the foundation for a new category: content management systems built for AI agents, not just human browsers.
Declare what agents can do with your content and services
Structured content optimized for agent consumption
Type-safe definitions for agent operations
Track behavior, measure value, optimize content
Identity, authentication, and billing per agent
Turn agent traffic into sustainable revenue
ROADMAP
A phased approach to universal agent access—from infrastructure to ecosystem.
Core specification, CLI tools, validators, and reference implementations.
Framework integrations, package distribution, and developer experience.
Agent analytics platform, OAuth tool framework, and monetization tools.
Standards adoption, multi-agent coordination, and full agentic commerce.
THE ECONOMIC CASE
NBER research shows AI agents reduce transaction costs toward near-zero. ARW enables this transformation by providing the infrastructure for efficient agent-web interaction.
"ARW is essential infrastructure for the agent economy, not an optional enhancement. As agent traffic grows from 40% to 70%, websites without ARW will be at a severe competitive disadvantage."
We built ARW as a collaborative, open source standard and we welcome your feedback on how to improve it. Whether you're a publisher protecting your content, an AI company building better agents, or a developer implementing the spec, your input helps shape the future of the agent web.
Contact us