Google Releases an Open Standard for AI Knowledge

Google Cloud published an open standard on 12 June 2026 called the Open Knowledge Format (OKF) — a draft specification for the internal context that AI agents need to do useful work.

The release formalises a pattern that has been spreading under different names for the past year: a folder of plain-text files that an AI agent can read, write and update on its own. Google is betting that the pattern needs a name, not a product.

What Google released

The Open Knowledge Format is published as v0.1 with draft status. It is vendor-neutral, agent- and human-friendly — a way to represent the metadata, context and curated knowledge that modern AI systems need. The specification is open and the source lives in the GoogleCloudPlatform/knowledge-catalog repository on GitHub. (Google Cloud blog)

The problem it solves is mundane but expensive. In most firms, the knowledge an agent would need to answer a sensible question — table schemas, metric definitions, an incident runbook, a deprecated API — lives in incompatible places: metadata catalogs with their own APIs, wikis, code comments, shared drives, the heads of a few senior engineers. Every agent builder solves the same context-assembly problem from scratch, and the knowledge stays locked inside whichever tool created it.

OKF is the suggestion that the answer is a format, not another service. The same way a Word file lets anyone with Word read your essay, OKF is meant to let any agent read your knowledge — without separate software to install, without a proprietary account, without sending the data to anyone.

The pattern it formalises

“LLMs don’t get bored, don’t forget to update a cross-reference, and can touch 15 files in one pass.”

That line is from AI researcher Andrej Karpathy’s LLM Wiki gist, and Google uses it to explain why a folder of plain-text files is interesting. The same idea now appears as Obsidian vaults wired to coding agents, convention files like AGENTS.md and CLAUDE.md, and “metadata as code” repositories inside data teams.

Each instance looks alike, but none were designed to cooperate. OKF puts a small set of conventions on top so a wiki written by one team can be read by an agent built somewhere else.

The spec is built on three principles:

• Minimally opinionated. Every concept needs a single type tag. Everything else is up to the producer.
• Producer/consumer independence. The same file is readable by a human in a text editor and by an AI agent. Tooling on either end can be swapped.
• Format, not platform. No specific cloud, database, model or framework is required.

Google shipped three reference pieces alongside the spec: an enrichment agent that walks a BigQuery dataset and drafts an OKF concept for every table, a static visualizer that renders any bundle as an interactive graph in a single self-contained file, and three sample bundles covering GA4 e-commerce, Stack Overflow and Bitcoin public datasets.

What to watch

Google is explicit that v0.1 is a starting point, designed for backward-compatible growth, and the format will evolve as more producers and consumers emerge. (Google Cloud blog)

For UK small firms, the practical answer is not yet — nothing crawls the web for OKF bundles, the standard is two weeks old, and Google built it for data teams. The cost of putting one down is asymmetric, though: a free generator turns a site into OKF in minutes, and the structural map is a useful audit even if no agent ever opens it. (Suganthan)

The strategic shift underneath matters more. Google is also rebranding Dataplex as Knowledge Catalog, described by Suganthan as an “always-on context engine” for AI agents, with OKF positioned as the open, portable piece. If a single vendor’s format takes hold, the question for UK buyers is whether the rest of the stack stays open, or whether the open standard quietly becomes lock-in. (Suganthan)

Three things to keep an eye on:

• Adoption beyond Google. The repository carries Google’s standard boilerplate for unsupported open-source code — a signal that the spec lives or dies on community use.
• Producers and consumers. The first generation of tools (enrichment agents, visualizers, search indexes) will determine how painful the format is to use.
• A web crawler’s behaviour. Until something reliably fetches OKF bundles, the format is a registration exercise, not an advertising one.

The standard is worth knowing about. The standard is not yet worth budgeting for.