Google Cloud Announces The Open Knowledge Format

Jun 15, 2026 05:07 PM - 1 month ago 36702

Google announced the Open Knowledge Format (OKF), a caller unfastened specification for organizing and exchanging the knowledge that AI systems request successful bid to execute useful work.

The announcement explains the logic for processing this caller specification:

“As instauration models proceed to improve, the deficiency of applicable discourse often limits what they tin do, particularly arsenic they are utilized to build agentic systems. While these models tin thief you constitute code, summarize documents, aliases analyse a dataset, they still request the correct accusation to nutrient meticulous and actionable results. “

AI Agents Need Context

AI systems often request knowledge that exists extracurricular the model, including really information is structured, really systems work, really metrics are defined, and really soul processes operate.

That knowledge is usually scattered crossed catalogs, wikis, documentation, repositories, shared drives, and different soul systems, forcing AI agents to combine discourse earlier they tin complete a task.

Google says OKF is meant to lick that problem by turning scattered knowledge sources into a communal format that tin move betwixt humans, AI agents, tools, and organizations.

What Open Knowledge Format Is

OKF is simply a format for representing organizational knowledge successful a measurement that tin beryllium shared betwixt different AI agents, tools, and organizations.

The format organizes concepts specified arsenic datasets, metrics, APIs, tables, and runbooks into documents that tin beryllium publication by some humans and AI systems.

Google designed OKF to beryllium elemental and independent of immoderate circumstantial platform, allowing the aforesaid knowledge to beryllium shared betwixt different AI agents, tools, and organizations.

The announcement explains:

“To make the format concrete, we’re publishing reference implementations astatine some the shaper and user ends:

An enrichment supplier that walks a BigQuery dataset, drafts an OKF conception archive for each array and view, past runs a 2nd LLM walk that crawls charismatic archiving and enriches each conception pinch citations, schemas, and subordinate paths.
A fixed HTML visualizer that turns immoderate OKF bundle into an interactive chart position successful a azygous self-contained file; nary backend, nary instal connected the viewing side, nary information leaves the page.
Three ready-to-browse sample bundles: GA4 e-commerce, Stack Overflow, and Bitcoin nationalist datasets, produced by the reference supplier and committed to the repo arsenic surviving examples of conformant OKF.

These are proofs of concept, deliberately. The supplier demonstrates 1 measurement to nutrient OKF; thing astir the format requires a circumstantial supplier model aliases LLM. The visualizer demonstrates 1 measurement to devour it; thing astir the format requires HTML aliases a chart view. We expect (and want!) the ecosystem of producers and consumers to turn acold beyond what we’ve shipped.”

Who OKF Is For

OKF is designed astir a producer-and-consumer model. Some users create, edit, and support the knowledge. Others devour it done AI agents, LLMs, package systems, aliases soul tools.

AI Agents and LLMs

AI agents and LLMs are the superior consumers of OKF. They usage the format to entree the system discourse and curated knowledge needed to execute tasks and nutrient meticulous results.

Useful For AI Agents And LLMs

Coding agents
Data study agents
Research agents
Internal endeavor assistants
Agentic workflows

Humans And OKF

OKF uses markdown files and YAML frontmatter, making the format readable and editable by group utilizing modular tools.

People Who May Find OKF Useful

AI developers
Software engineers
Data engineers
Analytics teams
Technical writers
Business teams

Organizations And OKF

Organizations tin usage OKF to package and stock organization knowledge that would different stay scattered crossed archiving systems, metadata catalogs, repositories, and soul tools.

Organizations That May Find OKF Useful

Organizations building AI agents
Data teams
Engineering teams
Knowledge guidance teams

Availability

Google is proposing a communal format for representing organizational knowledge alternatively than a caller level for managing it. The OKF specification, reference implementations, and sample bundles are available connected GitHub. The announcement makes a constituent of saying that it is simply a starting point:

“OKF v0.1 is simply a starting point, not a vanished standard. The format will germinate arsenic much producers and consumers look and arsenic we collectively study what knowledge representations agents really request successful practice.

We’re publishing successful the unfastened from time 1 because that’s the only measurement a knowledge format earns its name, whether you’re building a knowledge catalog, an enrichment pipeline, a wiki tailored to AI agents, aliases thing successful the AI knowledge domain.”

An explainer tweet by Tech With Mak shared why this solves a problem:

“The astir underrated thought successful supplier tooling this twelvemonth mightiness beryllium a gist Andrej Karpathy wrote astir “LLM Wikis” – markdown libraries that agents read, update, and support connected their own.

What followed was predictable. Teams everyplace started building their ain type – AGENTS[.]md, CLAUDE[.]md, Obsidian vaults wired into coding agents, folders of index[.]md and log[.]md files agents consult earlier doing anything.

…Google conscionable tried to adjacent that spread pinch the Open Knowledge Format – a spec that says => here’s the 1 section each conception needs (type), here’s a mini group of optional fields if you want them queryable, and otherwise, constitute nevertheless you want.

It’s not a caller instrumentality aliases platform. It’s an statement connected shape, which is precisely what Karpathy’s shape needed to extremity being a 100 incompatible reinventions of the aforesaid idea.”