Google's Open Knowledge Format (OKF) v0.1 — published on 2026-06-13 by Google's data team — is a directory of linked markdown files with YAML frontmatter that expresses a concept graph. No runtime, no SDK, no build step. Just markdown, just files, just YAML frontmatter. It is the format that flat websites need in order to be understood by AI agents, not just crawled.
The ciwebgroup.com knowledge graph is live at /okf/. 79 concept files across nine categories (organization, people, platform, agents, strategy pillars, services, products, concepts, research). Every relative link is a graph edge; every YAML block is a machine-readable definition.
OKF v0.1 is a spec you could implement in a weekend. Its power is not the technology; it is the discipline of publishing your knowledge graph in a format every AI agent already knows how to parse.
Each entity — a service, a product, a pillar, an agent, a person, a research finding — gets its own .md file. No SDK, no database, no build step. Just files.
Six fields on every file: type, title, description, resource (canonical URL back to the human page), tags (array), timestamp (ISO 8601). Predictable machine format an agent can parse without inference.
A pillar file links to its component files with a standard [text](../products/x.md). The links ARE the graph. No JSON-LD to write, no separate ontology file to maintain.
Every OKF bundle carries a top-level manifest listing every concept file. An AI agent fetches /okf/index.md first, then traverses the graph.
When a ChatGPT operator, Claude Computer Use session, or Perplexity browser traverses your site, it fetches one page at a time. Nothing on those pages tells it how the concepts relate. Two pages can mention the same concept without ever telling a machine that one is the framework and the other is a sub-goal.
A knowledge graph says it outright. An agent that fetches your OKF manifest first knows which pillar every product powers, which agents run which services, which research findings back which claims, and who authored what. That is the difference between being indexed and being understood.
OKF is not the first attempt to make websites machine-legible. The ones that failed all share a shape: they exposed a piece of the picture without exposing the relationships between the pieces.
The Central Intelligence Engine — vector memory, RAG, CAG, MemPalace, the CI Agents suite — is a knowledge graph. It has been running as one for years. What OKF adds is public, standard, agent-fetchable. Here is what ciwebgroup.com already shipped before OKF existed:
Every client's site already runs on a per-tenant vector store with retrieval- and cache-augmented generation. That is the internal graph OKF externalizes.
The memory layer alongside Vector, RAG, and CAG on the sitewide FrameworkBand. Long-lived structured relationships every CI Agent reads from and writes to.
Eight named agents coordinating via Model Context Protocol on top of shared memory. Wolverine, Groot, Neo, Sarge, Yoda, Buttercup, Goodspeed, Vader.
The same entities appear in the sitewide Organization schema, in llms.txt, in the featured-customer records, and in the release-note author line. OKF is just the export.
79 concept files under /okf/ at ciwebgroup.com — every service, every one of the 48 Hydra Build Components, every strategy pillar, every named CI Agent, both research stat sets, the CI Web Group organization node, and Jennifer Bagley as the canonical author entity.
The manifest at /okf/index.md is regenerated every time an underlying source module changes. The graph never drifts from the site.
Every product file links up to its strategy pillar, every pillar links up to the organization, every agent links to CI Agents (the suite), and every concept links laterally to the ones it touches. There are no dead edges.
The Jennifer Bagley person node is a first-class OKF entity. Every research finding is attributed. Every service links back to a signed E-E-A-T author. This is the trust layer OKF was built to make readable.
The bundle is regenerated from live TypeScript source (lib/build-components.ts, lib/team.ts, lib/platform.ts, lib/authority.ts) by scripts/export-okf.mjs. If the site changes, the graph regenerates. If the graph disagrees with the site, the site is the source of truth and the bundle regenerates on the next build.
The identity file — llms.txt — grows into a full knowledge graph. That is what an OKF bundle is: the graph version of what llms.txt has always been trying to communicate.
AI agents query the graph instead of scraping pages. Fetching /okf/index.md and traversing 20 relative links is faster, cheaper, and more accurate than crawling 400 pages of prose.
The knowledge map becomes the canonical layer. The machine-readable version IS the source; the human pages become one rendering of it — a marketing surface on top of the entity graph.
CI is already building this. The Central Intelligence Engine is a knowledge graph. OKF just made it publishable.
Talk to a real strategist — not a chatbot. We'll show you exactly how Hydra OS would work for your business, franchise system, portfolio, dealer network, or agency.
30 minutes with a senior strategist. Walk away with a Hydra OS plan for your service area.
Brand-consistent local websites, AI-search visibility, and reporting across every market.
Give contractors a direct path to Hydra OS through the brands and suppliers they already trust.
Resell blogging, web design, development & SEO under your brand. Margin without the headcount. Q4 2026.
Know a contractor who should be on Hydra OS? Refer them and earn AI credits when they join — unlimited referrals through the CI Web Group Ambassador Program.
Refer & earn creditsDozens of Hydra OS sites are in production right now — watch them get built in real time. See the live launches and work in progress in our portfolio.
See the work