A new open-source project called KOS Protocol wants to give AI agents a reliable source of verified facts about websites - and the proposed format is a kos.json file placed at a domain's root.
The concept draws from established web standards practice. robots.txt tells search crawlers which pages to skip. sitemap.xml guides them to indexable content. kos.json would serve a different purpose: letting site owners publish accurate information about their organization, products, or content that AI agents can read directly. When an AI agent is autonomously browsing the web to research a company - for a prospect report, a comparison tool, or a product recommendation - it frequently pulls from outdated cached data or assembles facts from unrelated sources. Getting company details wrong is one of the most common failure modes for AI agents doing research tasks.
The underlying problem is legitimate. Whether this specific format solves it depends almost entirely on whether major platforms choose to support reading kos.json files. The AI agent standards space is crowded right now - Anthropic's Model Context Protocol (MCP), various OpenAPI specifications, and others are competing to define how agents interact with web services. KOS Protocol has a narrower, simpler scope than most - verified facts only, not full service integration - which could be an advantage or could make it insufficient for developers who need more.
The project has minimal public traction as of April 2026. For anyone building AI agent workflows that require accurate, up-to-date company or product information, the core idea of a verified facts endpoint has real merit - regardless of whether this particular implementation gains adoption.