An open web standard for making any website understandable and actionable by autonomous AI agents.
robots.txttaught crawlers where to go.sitemap.xmlshowed search engines what exists. ARA tells AI agents what a site is, what it offers, and how to interact with it.
The web was built for humans who browse visually. Search engines brought standards like robots.txt, sitemap.xml, and Schema.org for indexing. But no standard exists for AI agents that need to understand, navigate, and interact with a website programmatically.
Today, an AI agent visiting a website must:
- Parse raw HTML/DOM (thousands of tokens of noise)
- Take screenshots and interpret them visually (slow, expensive)
- Guess at site structure, available actions, and data schemas
- Interact through fragile UI automation (clicking buttons, filling forms)
This costs 10-20x more tokens than necessary and produces unreliable results.
ARA provides a 3-layer architecture that lets any website declare itself "agent-ready":
┌─────────────────────────────────────────────────┐
│ LAYER 3 — INTERACTION │
│ Actions, MCP/A2A/REST Endpoints, Workflows │
│ /.well-known/ara/actions.json │
├─────────────────────────────────────────────────┤
│ LAYER 2 — UNDERSTANDING │
│ Semantic Schemas, Knowledge Graph │
│ /.well-known/ara/schemas/ │
├─────────────────────────────────────────────────┤
│ LAYER 1 — DISCOVERY │
│ Identity, Catalog, Content Map │
│ /.well-known/ara/manifest.json │
└─────────────────────────────────────────────────┘
Core principle: "Understand in 1 request." A single HTTP request to manifest.json gives an agent a complete understanding of any website.
A minimal ARA manifest for an e-commerce site:
{
"$ara": "1.0",
"identity": {
"name": "TechShop",
"type": "ecommerce",
"description": "Online electronics store with 2,000+ products",
"locale": ["en-US"]
},
"content_map": {
"summary": "2,000 products across 15 categories",
"resources": [
{
"id": "products",
"type": "catalog",
"label": "Product Catalog",
"count": 2000,
"schema_ref": "schemas/product.json",
"endpoint": "/ara/resources/products"
}
]
},
"capabilities": {
"protocols": {
"rest_api": {
"openapi": "https://techshop.com/api/openapi.json"
}
}
},
"policies": {
"agent_access": "open",
"rate_limit": { "requests_per_minute": 60 }
}
}That's ~30 lines of JSON — and an agent now knows what TechShop is, what it sells, how to query its catalog, and what the access rules are.
| Approach | Tokens to understand a site | Reliability |
|---|---|---|
| DOM/HTML parsing | 15,000 – 50,000 | Fragile |
| Screenshot analysis | 5,000 – 15,000 per page | Approximate |
| ARA manifest + digest | 500 – 1,500 | Structured & reliable |
ARA reduces agent token consumption by 10-20x.
| Capability | robots.txt | sitemap.xml | Schema.org | llms.txt | OpenAPI | ARA |
|---|---|---|---|---|---|---|
| Site discovery | — | Partial | — | Partial | — | Complete |
| Global overview | — | URLs only | — | Plain text | — | Structured |
| Data schemas | — | — | Fragmented | — | Yes | Semantic |
| Actions | — | — | Limited | — | Yes | Multi-protocol |
| Intent mapping | — | — | — | — | — | Native |
| MCP/A2A support | — | — | — | — | — | Native |
| LLM-optimized digest | — | — | — | Basic | — | Optimized |
| Agent policies | Basic | — | — | — | Partial | Complete |
| Incremental adoption | — | — | — | — | — | 4 levels |
You don't need to implement everything at once. ARA supports incremental adoption:
| Level | What you add | Effort | Agent benefit |
|---|---|---|---|
| Level 0 | Nothing | None | Agent must parse HTML (slow, fragile) |
| Level 1 | manifest.json only |
1 file | Discovery + global understanding |
| Level 2 | Manifest + Schemas | 3-5 files | Full structural understanding |
| Level 3 | Manifest + Actions | 5-10 files | Programmatic interaction |
| Level 4 | Full ARA + MCP/A2A | Complete integration | Native agent experience |
Even Level 1 (a single JSON file) reduces token consumption by 90%.
- Create the directory:
mkdir -p .well-known/ara- Create your
manifest.json— start from one of our examples:
cp examples/minimal-manifest.json .well-known/ara/manifest.json
# Edit with your site's details- Validate:
npx ara-validate https://yoursite.comimport requests
# One request to understand any ARA-ready site
manifest = requests.get("https://example.com/.well-known/ara/manifest.json").json()
print(manifest["identity"]["name"]) # "TechShop"
print(manifest["identity"]["type"]) # "ecommerce"
print(manifest["content_map"]["summary"]) # "2,000 products across 15 categories"
# Access resources via declared protocols
for resource in manifest["content_map"]["resources"]:
print(f"{resource['label']}: {resource['count']} items")- Full Specification (v1.0) — Complete technical specification
- Examples — Ready-to-use manifests for different site types
- Tools — Validator and generator
- E-commerce: An agent compares products across 15 ARA-ready stores in seconds, without parsing a single HTML page.
- SaaS: An enterprise agent evaluates CRM solutions by reading manifests, comparing features via schemas, and simulating pricing via actions.
- Content/Media: A monitoring agent ingests 500 news sites via Content Digests (~300 tokens each) and produces a daily synthesis.
- Local business: A personal assistant finds a restaurant, reads the menu, and books a table — all through structured ARA interactions.
| Phase | Timeline | Goal |
|---|---|---|
| Phase 1 — Specification | Q1 2026 | Publish ARA v1.0 spec |
| Phase 2 — Tooling | Q2 2026 | CLI, CMS plugins (WordPress, Shopify), validator |
| Phase 3 — Early adoption | Q3 2026 | 100 pilot sites, partnerships with 3 agent frameworks |
| Phase 4 — Ecosystem | Q4 2026 | Multi-language SDKs, agent registry, certification |
| Phase 5 — Standardization | 2027 | W3C / IETF submission |
We welcome contributions! See CONTRIBUTING.md for guidelines.
- Report bugs and suggest features via Issues
- Submit improvements via Pull Requests
- Discuss ideas in Discussions
This specification is released under Creative Commons Attribution 4.0 International (CC BY 4.0).
The agentic web is here. ARA gives every website a front door for AI agents.
Website · Specification · Examples · Discuss