Specification Registry

Single source of truth for every AI Discovery File defined by this specification.

Status Stable

The registry is published and current. Downstream consumers fetch /specifications/registry.json to discover the canonical list of files, their media types, paired files, precedence rules, and schemas.

Last updated:

Purpose

The registry is the machine-readable index of every AI Discovery File. Validators, plugins, and other downstream consumers SHOULD load /specifications/registry.json at runtime rather than hard-coding the file list. The same data is rendered here for human browsing.

Machine-readable formats

Files in the registry

Registry version 1.5.0, dated 2026-05-11. Each entry below lists the file's identifier, its path, its media types, its current version, the conformance classes it participates in, and any paired files or precedence rules.

ID File Path Media types Version Conformance classes
ADF-001 llms.txt /llms.txt text/markdown; charset=utf-8
text/plain; charset=utf-8 		1.7.0
ADF-002 llm.txt /llm.txt text/plain; charset=utf-8 1.7.0
ADF-003 llms.html /llms.html text/html; charset=utf-8 1.7.0
ADF-004 ai.txt /ai.txt text/plain; charset=utf-8 1.7.0
ADF-005 ai.json /ai.json application/json 1.7.0
ADF-006 identity.json /identity.json application/json 1.7.0
ADF-007 brand.txt /brand.txt text/plain; charset=utf-8 1.7.0
ADF-008 faq-ai.txt /faq-ai.txt text/plain; charset=utf-8 1.8.0
ADF-009 developer-ai.txt /developer-ai.txt text/plain; charset=utf-8 1.8.0
ADF-010 robots-ai.txt /robots-ai.txt text/plain; charset=utf-8 1.7.0

Precedence relationships

Where two or more files address the same field, the precedence relationships below apply. For the full discussion, see the Interoperability Guide.

llms.txt
  • vs llm.txt: llms.txt is canonical; llm.txt SHOULD be a 301 redirect to llms.txt.
  • vs llms.html: Same content, different format; llms.html is the human-readable companion to llms.txt.
llm.txt
  • vs llms.txt: llm.txt SHOULD return HTTP 301 to /llms.txt; it MUST NOT carry independent content.
llms.html
  • vs llms.txt: llms.html mirrors llms.txt content in HTML; structured-data takes precedence over unstructured if they disagree.
ai.txt
  • vs ai.json: ai.json is the structured, parseable form and takes precedence over ai.txt if they disagree.
  • vs robots.txt: robots.txt is an external standard and always takes precedence for access control.
ai.json
  • vs ai.txt: ai.json is the canonical, machine-parseable form; ai.txt is the human-readable summary.
  • vs robots.txt: robots.txt is an external standard and always takes precedence for access control.
identity.json
  • vs llms.txt: identity.json is authoritative for structured identity; llms.txt provides narrative context.
  • vs brand.txt: brand.txt is authoritative for naming and pronunciation; identity.json carries the legal and operational identity.
brand.txt
  • vs identity.json: brand.txt is authoritative for naming, pronunciation, and terminology guidance.
robots-ai.txt
  • vs robots.txt: robots.txt is an external standard and always takes precedence; robots-ai.txt supplements with AI-specific directives but MUST NOT contradict it.

JSON Schemas in the registry

Two files in the registry have published JSON Schemas. Publishers SHOULD reference these via $schema in their own files so editors can autocomplete and validators can detect malformed input.

Consuming the registry

A typical consumer fetches /specifications/registry.json, iterates the files array, and uses each entry's path to construct the file URL on the target site. The conformanceClasses array tells the consumer which class membership each file participates in; see the Conformance specification for the formal rules.

The reference validator (the AI Visibility Checker) emits output conforming to validator-output.schema.json. Other conformant validators SHOULD do the same so downstream tooling can consume validator output portably.

References