Google Cloud launches Open Knowledge Format for AI agents

A new specification from Google Cloud is trying to do something modest but useful, agree on what a folder of context for an AI should look like. In a Google Cloud blog post, tech leads Sam McVeety and Amir Hormati introduced the Open Knowledge Format v0.1, described in the post as a vendor-neutral, agent- and human-friendly standard for representing the metadata, context, and curated knowledge that modern AI systems need.

The format itself is deliberately boring. An OKF bundle is a directory of markdown files with YAML frontmatter holding a small set of fields like type, title, description, resource, tags, and timestamp. The framing in the post is just markdown, just files, just YAML frontmatter, readable in any editor, renderable on GitHub, hostable in any git repo. The design principle the authors emphasize is format, not platform, with the claim that it will never require a proprietary account or SDK to read, write, or serve.

The reason that framing matters is that the problem OKF targets is genuinely fragmented. The post argues that internal knowledge today lives in metadata catalogs with proprietary APIs, wikis, code comments, and as it puts it, the heads of senior engineers. The diagnosis is that every agent builder is solving the same context-assembly problem from scratch. The pitch is to standardize the shape of that context so a bundle written by a human can be consumed by an agent, and a bundle exported from a metadata pipeline can be browsed in a visualizer.

Google is shipping reference implementations alongside the spec, an Enrichment Agent that walks a BigQuery dataset and drafts OKF documents for every table and view, a static HTML visualizer that turns a bundle into an interactive graph, and three sample bundles built on the GA4 e-commerce, Stack Overflow, and Bitcoin public datasets. Google has also updated its own Knowledge Catalog to ingest OKF and serve it to agents. The lineage the post calls out is Andrej Karpathy's LLM Wiki gist, quoted as "LLMs don't get bored, don't forget to update a cross-reference, and can touch 15 files in one pass," alongside Obsidian vaults wired to coding agents and AGENTS.md / CLAUDE.md convention files.

The honest caveat is that a vendor publishing an open spec is not the same thing as the industry adopting it, and the post itself calls v0.1 a starting point, not a finished standard. What the announcement does not tell you is how OKF will be governed beyond Google, whether other catalog vendors or cloud providers plan to write producers, or how it lines up against the AGENTS.md style conventions Google itself cites as influences. If those questions go the right way, the upside is straightforward, small teams get a portable way to ship the context their agents need, and tool builders finally get one target to write against.