The Open Knowledge Format: AI-ready knowledge without lock-in

Most organisations sit on a surprising amount of knowledge: runbooks, architecture decisions, product specs, support answers, the hard-won context that lives in a handful of people's heads. The moment you try to put AI to work on top of that knowledge — a support assistant, an internal search tool, a retrieval-augmented chatbot — you run into the same wall. The knowledge isn't in a form a model can use, and whatever form you do put it in tends to get locked inside one vendor's platform.

The Open Knowledge Format (OKF), a vendor-neutral specification published by Google, is a quietly important answer to that problem. It's worth understanding now, because the way you structure your knowledge today determines how easily you can adopt — and switch between — AI tools tomorrow.

What OKF actually is

OKF is deliberately unglamorous. A "bundle" of knowledge is just a directory of plain Markdown files, each with a small block of YAML metadata at the top (a type, a title, a few tags, a timestamp). Concepts link to each other with ordinary Markdown links, which together form a knowledge graph. Sources and citations are tracked so claims can be traced back to where they came from. That's essentially the whole format.

The power is in what that simplicity buys you:

• It's portable. Plain Markdown and YAML are readable by humans, by every AI model, and by any tool that can read a text file. Nothing proprietary.
• It's version-controlled. Because a bundle is just files, it lives in Git like the rest of your engineering work — with history, review, and diffs.
• It mixes structure and prose. The metadata gives machines something reliable to parse; the Markdown body keeps it readable for people.
• It's vendor-neutral. No platform owns it. You can feed the same bundle to different AI systems, or move between them, without re-exporting and reformatting everything.

If that sounds a lot like a well-organised wiki, that's the point. OKF takes a pattern many teams already reach for and makes it a standard that tools can agree on.

Why this matters for AI readiness

Ask why an AI initiative stalled and the answer is rarely the model. It's almost always the data and the knowledge feeding it. Retrieval-augmented generation (RAG) — the technique behind most useful internal AI assistants — is only as good as the knowledge base it retrieves from. If that knowledge is scattered across wikis, slide decks, ticket threads, and inboxes, the assistant's answers will be scattered too.

Getting your knowledge into a structured, portable form is one of the highest-leverage things you can do to become genuinely AI-ready. It pays off in three ways:

1. Better grounding. A clean, linked, source-cited knowledge base gives an AI system something trustworthy to draw on, which directly reduces the confident-but-wrong answers that erode user trust.
2. No lock-in. Models and platforms are moving fast. If your knowledge lives in a neutral format, adopting a new tool — or replacing one that didn't work out — is a configuration change, not a migration project.
3. Provenance you can defend. OKF's citation convention means every claim can point back to a source. For regulated or risk-conscious organisations, being able to answer "where did the AI get that?" is not optional.

There's also a strategic angle. The format is genuinely new, and the organisations that treat their knowledge as a first-class, portable asset will compound that advantage as the tooling around standards like OKF matures.

How it fits a real architecture

A common and sensible pattern is to keep your knowledge in a database or wiki as the working source of truth, and treat OKF as the interchange layer — the portable form you export to and import from. Your knowledge stays where your team edits it; OKF is how it travels: into a Git repository, into an AI grounding pipeline, into the next tool you trial.

That separation matters. You don't have to rebuild how your team captures knowledge to get the benefits. You add a clean, standard way for that knowledge to leave the building and feed your AI systems — and a clean way for external knowledge bundles to come in.

Where to start

You don't need a large programme to begin. A pragmatic first step is to take one well-bounded domain — a product area, a set of runbooks, a body of support knowledge — and structure it as a bundle. A few things make the difference between a bundle that helps and one that gathers dust:

• Favour structure over prose. Headings, lists, tables, and short concept pages are easier for both people and models to use than long, unbroken documents.
• Link concepts together. The connections between ideas are part of the knowledge. A graph of linked concepts is far more useful to a retrieval system than a pile of isolated pages.
• Cite sources. Tie concrete claims back to where they came from. It builds trust and makes the knowledge auditable.
• Keep it close to your engineering workflow. Knowledge in Git, reviewed like code, stays current. Knowledge in a forgotten folder rots.

Turning raw organisational knowledge into a reliable AI asset touches several disciplines at once. Our work in AI Engineering covers the retrieval pipelines and embedding strategies that make formats like OKF performant in production, while Data Infrastructure addresses the storage and versioning decisions that keep knowledge portable over time. If you're still working out where AI fits in your broader product or platform roadmap, AI Product Strategy is a good place to start.

The takeaway

The Open Knowledge Format isn't a product to buy or a platform to adopt. It's a sensible, open standard for treating your organisation's knowledge as a portable, AI-ready asset rather than something trapped in a tool. As AI becomes a standard part of how organisations operate, the ones that have structured their knowledge deliberately — and kept it free of lock-in — will move faster and switch tools more freely than the ones still wrestling with scattered documents.

If you're weighing where to invest to get more out of AI, structuring your knowledge is rarely the most exciting line item on the roadmap. It's often the one that makes everything after it work.

---

If you're trying to make your organisation's knowledge genuinely AI-ready — without betting on a single vendor's format — we're happy to talk through what that looks like in practice. Get in touch and we can work through the specifics with you.