Google’s Open Knowledge Format: New Standard for AI Agents

Picture this: you’re a capybara building an AI agent in the swamp of fragmented knowledge. Your agent needs to know how to compute weekly active users, but the answer is scattered across a metadata catalog, a wiki, a senior engineer’s brain, and a dusty shared drive. Sound familiar? Google Cloud just threw a lifeline: the Open Knowledge Format (OKF), an open spec that turns your messy knowledge swamp into a clear, navigable channel.

What’s the Big Deal?

OKF is a vendor-neutral, agent-friendly format for representing metadata, context, and curated knowledge. Think of it as the lingua franca for AI agents—a way to package knowledge as a directory of markdown files with YAML frontmatter. No SDKs, no proprietary runtimes, just files. It’s like the Neon Database of knowledge formats: simple, open, and built for the modern data stack.

Why Now?

Foundation models are getting smarter, but they still need the right context to produce accurate results. As Andrej Karpathy famously noted in his LLM Wiki gist, LLMs don’t get bored updating cross-references—they thrive on the bookkeeping that makes humans abandon wikis. OKF formalizes this pattern into a portable, interoperable standard.

How It Works

An OKF bundle is a directory of markdown files, each representing a concept (table, dataset, metric, runbook). Each file has YAML frontmatter for structured fields like type, title, description, resource, tags, and timestamp, plus a markdown body for everything else. Concepts link to each other with normal markdown links, turning the directory into a rich graph of relationships.

sales/
├── index.md
├── datasets/
│   ├── index.md
│   └── orders_db.md
├── tables/
│   ├── index.md
│   ├── orders.md
│   └── customers.md
└── metrics/
    ├── index.md
    └── weekly_active_users.md

Here’s what a concept document looks like:

---
type: BigQuery Table
title: Orders
description: One row per completed customer order.
resource: https://console.cloud.google.com/bigquery?p=acme&d=sales&t=orders
tags: [sales, revenue]
timestamp: 2026-05-28T14:30:00Z
---

# Schema

| Column        | Type      | Description                              |
|---------------|-----------|------------------------------------------|
| `order_id`    | STRING    | Globally unique order identifier.        |
| `customer_id` | STRING    | FK to [customers](/tables/customers.md). |

# Joins

Joined with [customers](/tables/customers.md) on `customer_id`.

Three Guiding Principles

1. Minimally opinionated: Only requires a type field. Everything else is up to you.
2. Producer/consumer independence: A human-authored bundle can be consumed by an AI agent, and vice versa.
3. Format, not platform: No vendor lock-in. No proprietary accounts. Just open files.

What’s Shipping Today?

Google Cloud is releasing 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 any OKF bundle into an interactive graph view.
• Three sample bundles for GA4 e-commerce, Stack Overflow, and Bitcoin public datasets.

These are proofs of concept, but they show the power of a common format. The Google Cloud Knowledge Catalog has already been updated to ingest OKF, and the spec is open for contributions from the community.

Why This Matters for Capybaras

In the swamp of AI development, caimans (read: expensive proprietary solutions) have been hoarding knowledge in walled gardens. OKF is the capybara’s way: open, friendly, and built for collaboration. Whether you’re using Supabase, Firebase, or Vercel, OKF can be the bridge that connects your knowledge across tools.

Get Started

Head over to the GitHub repo to read the spec, try the reference implementation, or contribute. The format is designed to evolve with the community. As the swamp gets deeper, OKF keeps your knowledge flowing.

Original announcement published on Google Cloud.