What is FRED?
What it is, why it exists, and how it works. No jargon.
The Problem
AI agents are becoming the main interface between people and the internet. They browse, purchase, schedule, and research on your behalf.
But when an agent visits a website today, it faces a wall. There are 12+ different discovery protocols it needs to check — MCP, A2A, OpenAPI, llms.txt, Schema.org, robots.txt, and more. Each with their own file format, location, and setup process.
Site owners have to implement each one separately. Agents have to check each one separately. The agentic web is arriving — but it has no unified map.
The Solution
One JSON file. One well-known URL. Full story.
Any entity — a business, a person, an API, a community, a product — adds a fred.json file to their domain. FRED binds all existing protocols into a single entry point. AI agents read one file and know everything.
The 8 Layers
Only the first layer (Identity) is required. Everything else is optional and progressive.
| Layer | Question it answers | Example |
|---|---|---|
| Identity | Who are you? | Name, type, URL, contacts |
| Capability | What can you do? | Create CVs, search products, send emails |
| Interaction | How do I talk to you? | REST API, MCP server, GraphQL |
| Trust | Why should I trust you? | Verified business, 4.7★, 99.95% uptime |
| Pricing | What does it cost? | Free / $12/mo Pro / Enterprise |
| Context | What should I know? | FAQs, limitations, SLA |
| Discovery | Who is related? | Partners, integrations, alternatives |
| Policy | What are the rules? | Rate limits, data rights, attribution |
Protocol Map
FRED doesn't replace any of these protocols. It references them — telling agents where to find each one.
| Protocol | What it does | How FRED references it |
|---|---|---|
| MCP | Tool calling for AI agents | interaction.protocols[].type: "mcp" |
| A2A | Agent-to-agent communication | interaction.protocols[].type: "a2a" |
| OpenAPI | Machine-readable REST API contracts | interaction.protocols[].type: "openapi" |
| llms.txt | Structured context files for LLMs | context.llms_txt → URL to your llms.txt |
| ADP | Agent deployment and lifecycle | capabilities[] describes deployed agents |
| Schema.org | Semantic markup for search engines | Complementary — FRED is agent-first, Schema.org is crawler-first |
| robots.txt | Crawler access rules | policy.agent_policy extends this for AI agents |
| security.txt | Security disclosure contacts | trust.security.security_txt → URL |
| WebFinger | Resource and person discovery | Complementary identity discovery standard |
| JSON-LD | Linked data in JSON format | Complementary — FRED uses plain JSON, not RDF |
| Sitemap.xml | Content index for crawlers | context.knowledge_base can point to sitemaps |
| DNS TXT | Domain metadata via DNS | _fred.domain TXT records for discovery without HTTP |
Why It Matters
The web standardized on HTML. Email standardized on SMTP. The agentic web needs a standard too.
FRED is not a product. It's infrastructure. Open source, Apache 2.0, community-governed. Anyone can implement it. No account, no API key, no dependency on any company.
Every entity that adds fred.json makes the agentic web a little more navigable for every agent and every user.