The Open Knowledge Format is the easy part.
Google Cloud just published an open standard for the knowledge AI runs on. It validates a bet we shipped a year ago — and clarifies where the real value sits: not the format, the engine that fills and keeps it.
The context an AI needs to be useful is almost never in the model. It is your knowledge: the schema of a table, what a metric actually means in your business, the runbook for an incident, the reason a system was built one way and not the three others you ruled out. In most organizations that knowledge is scattered — across metadata catalogs, wikis, shared drives, code comments, and the heads of a few senior people. So every new AI tool re-solves the same context-assembly problem from scratch, and the knowledge stays locked behind whatever system happened to create it. On June 12, 2026, Google Cloud published an open specification aimed squarely at that problem: the Open Knowledge Format.
What OKF actually is
No new runtime, no SDK, no account to sign up for. A bundle of OKF knowledge is just markdown files with a little YAML frontmatter, arranged in a directory you can drop into a git repo. Readable by a person in any editor. Parseable by an agent with no translation layer. Diffable in version control. Portable across tools, organizations, and time. It takes the “LLM-wiki” pattern Andrej Karpathy described — a shared markdown library your agents read and maintain — and formalizes it into something many systems can speak.
The value of a knowledge format comes from how many parties speak it, not from who owns it.
Why a format, and not another platform
Most “AI that learns your business” keeps what it learns inside its own platform. The day you change tools or the contract ends, the learning leaves with it. A portable, vendor-neutral format inverts that: the knowledge bundle becomes the asset you own, and the tooling at each end becomes swappable. For a government program where staff rotate and contractors turn over — or a small business where one person quietly is the institutional memory — that is the difference between an asset that compounds and a silo you rent. A standard for portable knowledge is genuinely good news, and we are glad to see a major cloud put its weight behind one.
The part a format can’t solve
Here is the honest part. A standard for how to write knowledge down does not write the knowledge. OKF gives you the container; it cannot fill it. The expensive work is producing knowledge worth keeping and curating it while the work is happening — distilling the reasoning instead of dumping the transcript, encoding a failure as a standing rule so it cannot recur, keeping the cross-references current as things change. That upkeep is exactly why people abandon wikis. What makes the pattern finally viable is that the upkeep is now automatable. As Karpathy put it: “LLMs don’t get bored, don’t forget to update a cross-reference, and can touch 15 files in one pass. The bookkeeping that causes humans to abandon personal wikis is exactly what LLMs are good at.”
A standard tells you how to store knowledge. It does not tell you which knowledge is worth keeping.
What we have been building
We have run this exact shape in production for over a year: a structured knowledge base our agents build as a byproduct of the actual work — kept as plain markdown you own, on your infrastructure, with an audit trail for every entry. OKF did not change our architecture; it named it, and put an industry standard behind the bet. Google says the format is the contribution. For the person deciding what to buy, the format is the easy part — the engine that produces and curates the knowledge is the work, and that engine is what we sell. The portability OKF promises is the anti-lock-in guarantee we already make: your knowledge is an open, diffable bundle you can pick up and take anywhere.
Where we stand
A format gets more valuable the more parties speak it, so a shared, vendor-neutral standard is good for everyone building serious AI — us included. We will track OKF as it matures past its first draft. And we will keep optimizing for the property that mattered before there was a standard and matters more now that there is one: when the project ends or the tool changes, you keep not just the artifacts but the reasoning — in a form you own, that anyone can read.
Where the value actually sits.
OKF standardizes the container. The architecture page is how we fill and keep it; the sibling arguments are why the memory must be yours and why a vendor’s promise is weaker than one your own network can verify.