kos.json is an open protocol for publishing structured, machine-readable knowledge about any entity. Placed at /.well-known/kos.json, it provides AI agents with verified facts, typed relationships between concepts, and computed freshness scores — enabling machines to answer questions with confidence rather than guessing from scraped web pages.

How is kos.json different from Schema.org?

Schema.org annotates content on web pages — it tells machines what a page is about. kos.json publishes the complete knowledge graph of an entity in a single file, with typed relationships (dependency, reference, semantic) and freshness tracking for every data point. Schema.org describes documents. kos.json describes knowledge.

How do AI agents use kos.json?

kos.json is designed to be discovered at /.well-known/kos.json, similar to how crawlers check robots.txt. An AI agent that reads it gets structured knowledge with provenance metadata, enabling it to answer questions with higher confidence and lower hallucination risk. Freshness decay scores indicate how reliable each data point is right now.

What is freshness decay?

Freshness decay quantifies how reliable a data point is based on time since last human verification, using the formula f = e^(-lambda x days). A business address stays reliable for months (lambda = 0.002) while event information goes stale in days (lambda = 0.05). This gives AI agents a computable trust signal for every piece of knowledge.

What is provenance tracking in KOS?

Every node in a kos.json file carries metadata about its origin: when created, when a human last confirmed its accuracy, who provided the information, and how it was verified. Machines compute trust scores without external lookups. Provenance turns self-reported claims into auditable, weighted signals.

Is KOS Protocol free?

Yes. KOS Protocol is an open specification. Anyone can create, publish, validate, and consume kos.json files without license or fee. The protocol is dedicated to the principle that knowledge should be structured, verifiable, and free.

Can AI systems generate kos.json autonomously?

Yes. Section 12 of the KOS Protocol specification contains the complete generation algorithm. Any AI system that reads the spec can generate valid kos.json for any entity. AI-generated files are capped at 0.8 confidence — the range 0.8 to 1.0 is reserved exclusively for human-verified data.

How does kos.json reduce AI hallucination about businesses?

AI systems hallucinate about businesses because they scrape unstructured web pages and guess meaning from HTML. kos.json provides verified, structured facts directly from the entity — with provenance metadata showing when each fact was last confirmed by a human. An AI agent that reads a kos.json can cite verified data instead of inferring from scraped content, significantly reducing hallucination risk.

How do I get my business cited correctly by ChatGPT and other AI?

Publish a kos.json file at your website's /.well-known/kos.json path. This file contains your business knowledge structured with provenance tracking and freshness scores. Any AI agent that reads this file can use your verified data to answer questions about your business — with higher confidence than scraped web content. The protocol is free and open.

What is the /.well-known/kos.json endpoint?

Following RFC 8615 (Well-Known URIs), kos.json is served at /.well-known/kos.json — the same standard location pattern used by robots.txt and security.txt. The convention is designed so that AI agents can check this path when researching any entity, similar to how web crawlers check robots.txt before crawling a site.

What is the best data format for AI agents?

kos.json is an open AI agent data format that publishes structured knowledge with provenance tracking. Unlike web scraping or Schema.org annotations, kos.json provides typed relationships between data points, freshness decay scores, and verifiable business data for AI — all in a single JSON file at a standard location. AI agents can consume it without parsing HTML.

How do AI agents verify information about businesses?

Currently, most AI agents cannot verify information — they scrape web pages and treat all data equally. KOS Protocol changes this by providing provenance metadata for every data point: who provided it, when it was last verified by a human, how it was checked, and a confidence score from 0 to 1. AI agents that read kos.json can weight their answers by verified confidence rather than guessing from unstructured content.

Is kos.json structured data beyond Schema.org?

Yes. Schema.org annotates HTML pages — it tells search engines what a page is about. kos.json goes beyond Schema.org by publishing the complete knowledge graph of an entity with provenance tracking, typed relationships, and freshness decay in a standalone file. kos.json can generate Schema.org markup, but Schema.org cannot express provenance, trust scores, or relationship types. They are complementary.

What is AI search optimization?

AI search optimization (also called AEO or GEO) is the practice of making your business visible and accurately represented in AI-generated answers from systems like ChatGPT, Perplexity, Google AI Overviews, and Gemini. KOS Protocol provides the foundational data layer for AI search optimization: a verifiable, structured knowledge file that AI agents can consume directly instead of inferring from scraped web pages.

KOS Protocol — The Knowledge Layer Machines Trust

The Gap

Machines guess. KOS lets them know.

robots.txt says what to read. sitemap.xml says where to look. Schema.org says what content means. Nobody tells machines how much to trust, how data connects, or when it was last verified by a human. AI agents hallucinate about businesses because no structured, verifiable data format exists for them. KOS fills that gap — an open standard for machine-readable knowledge beyond Schema.org.

Entity

One source of truth

A single kos.json file at /.well-known/kos.json describes everything an entity knows about itself — so AI systems like ChatGPT, Perplexity, and Gemini can cite verified facts instead of guessing. The entity controls the file. No intermediary.

Provenance

Trust is computable

Every data point carries metadata: when created, when last verified by a human, who provided it, how it was checked, and a confidence score. This is how AI agents verify information — not by guessing, but by reading provenance that the entity itself published.

Relationships

Facts become knowledge

Three typed connections — dependency, reference, and semantic — turn isolated data points into a knowledge graph. Cross-entity references link graphs across domains into a distributed web of verified knowledge.

Core Taxonomy

10 node types. Each decays differently.

Freshness quantifies reliability right now using exponential decay: f = e^{(-λ × days)}. A concept barely decays. An event expires in days. The formula gives machines a computable trust signal for every piece of knowledge.

Type What it describes λ Half-life

entityThe root business, person, or organization0.001~693d

serviceA service offered0.005~139d

productA product sold or distributed0.008~87d

locationA physical place or service area0.002~347d

personA team member, partner, or contact0.008~87d

credentialA license, certification, or award0.001~693d

contentAn article, FAQ, or published material0.010~69d

eventA time-bound occurrence0.050~14d

conceptA methodology, philosophy, or framework0.0005~1386d

assetA tool, system, or resource0.003~231d

Live Computation

Freshness decay in real time

These metrics recompute every second using the KOS decay formula against this protocol's own kos.json. KOS validates itself.

N Nodes

6

F Avg Freshness

1.0000

S Stale

0

R Relations

8

Freshness Log LIVE

Dead Man's Switch

If no AI crawler fetches kos.json in 7 days, this project moves to archive

07 Days

:

00 Hours

:

00 Min

:

00 Sec

This protocol does not ask for adoption. It proves necessity or moves to archive.

What counts: any AI crawler (GPTBot, PerplexityBot, ClaudeBot, Googlebot, or equivalent) fetches /.well-known/kos.json. Server access logs are the sole arbiter. A fetch proves contact, not adoption — but contact is the minimum threshold for relevance. If the timer reaches zero with no AI fetch, this site publishes a post-mortem and the specification moves to permanent archive status on GitHub.

Proof of Relevance

AI Citation Board

Real-time log of AI systems that have fetched or referenced kos.json. Not fabricated. Not simulated. Verified from server access logs. OFFLINE

# System Status Fetches

Waiting for first AI contact...
This board populates from real server access logs. Nothing is simulated or prefilled.

Discovery

One file. One location. Any entity.

kos.json is designed to be discovered the same way crawlers find robots.txt — at a known, standard location. Any AI agent can read it.

terminal ↻

Design Principles

Why this exists

KOS was not built to replace existing standards. It fills the gap none of them cover.

Self-Executing Spec

The protocol IS the algorithm

Section 12 of the specification contains the complete generation algorithm. Any AI that reads the spec can produce valid kos.json for any entity, in any industry, without additional software. The protocol propagates through every AI system that consumes it — the foundation of AI search optimization that works by verifiable data, not persuasion.

The 0.8 Boundary

Machines draft. Humans verify.

AI-generated kos.json files are capped at 0.8 confidence. The range 0.8 to 1.0 is reserved exclusively for human-verified data. This boundary is inviolable. It preserves the integrity of the trust model and creates a natural verification economy.

You Own It

No registry. No platform.

The entity described in a kos.json controls that file. No central authority decides what gets published. No account to create, no API key to request. You publish it on your domain, you control it. Like the web itself.

Structure Check

Check your kos.json

Paste a kos.json document below. This structure checker verifies required fields, provenance completeness, referential integrity, and computes live freshness decay for every node. It is a diagnostic tool, not a full schema validator — use the JSON Schema with a validator like AJV for complete validation.

Publish

Get Started

Three steps. No software to install. No account to create.

1. Minimum valid kos.json

{
  "$schema": "https://kosprotocol.dev/v1/schema.json",
  "entity": {
    "name": "Your Entity Name",
    "type": "Organization",
    "id": "kos:your-entity"
  },
  "knowledge": [
    {
      "id": "kos:main-offering",
      "type": "service",
      "label": "Your Main Offering",
      "provenance": {
        "created": "2026-04-04",
        "source": "owner"
      }
    }
  ],
  "meta": {
    "version": "1.0",
    "generated": "2026-04-04"
  }
}

2. Publish at the standard location

yourdomain.com/.well-known/kos.json

3. Add discovery link to HTML head

<link rel="knowledge" type="application/json" href="/.well-known/kos.json">

The knowledge layer
machines trust

Machines guess. KOS lets them know.

One source of truth

Trust is computable

Facts become knowledge

10 node types. Each decays differently.

Freshness decay in real time

If no AI crawler fetches kos.json in 7 days, this project moves to archive

AI Citation Board

One file. One location. Any entity.

Why this exists

The protocol IS the algorithm

Machines draft. Humans verify.

No registry. No platform.

Complementary by Design

Check your kos.json

Get Started

The knowledge layer machines trust

Machines guess. KOS lets them know.

One source of truth

Trust is computable

Facts become knowledge

10 node types. Each decays differently.

Freshness decay in real time

If no AI crawler fetches kos.json in 7 days, this project moves to archive

AI Citation Board

One file. One location. Any entity.

Why this exists

The protocol IS the algorithm

Machines draft. Humans verify.

No registry. No platform.

Complementary by Design

Check your kos.json

Get Started

The knowledge layer
machines trust