Google OKF Spec Standardizes Markdown Context for AI Agents

Google Cloud released the Open Knowledge Format (OKF) to standardize how developers package organizational context for AI agents. The v0.1 specification formalizes the common pattern of storing agent instructions in Markdown files. By defining a consistent directory structure and metadata schema, OKF provides a vendor-neutral way to pass database schemas, business metrics, and runbooks across different AI tools without custom integration code.

Technical Specification

The format prioritizes human readability and requires no dedicated SDK or runtime. An OKF bundle is a directory of UTF-8 encoded .md files. This lightweight approach mirrors existing community practices while adding the formal structure required for enterprise systems.

Each file must include a YAML frontmatter block. The specification uses this metadata to help agents index and categorize the raw text.

Field	Requirement	Purpose/Example
type	Required	Categorizes the content (e.g., table, metric, process)
title	Optional	Human-readable document name
description	Optional	Brief summary of the file contents
resource	Optional	URL pointing to the original data source
tags	Optional	Keywords for filtering and retrieval
timestamp	Optional	Tracks document freshness

The specification relies on standard Markdown links for navigation between documents. Google reserved specific filenames within the spec, designating index.md for navigation lists and log.md for change history tracking. The entire v0.1 specification is available on GitHub under the Apache 2.0 license.

Tooling and Catalog Integration

Alongside the spec, Google released reference implementations to help developers generate and review these bundles. An Enrichment Agent can scan BigQuery datasets and automatically draft OKF documents mapping tables and views. This automates the ingestion of structured data into text formats that AI coding assistants can readily process.

A Static HTML Visualizer tool converts OKF bundles into interactive graphs, allowing human reviewers to audit the context before feeding it to an agent. Google also provided sample datasets covering Google Analytics 4 e-commerce, Stack Overflow, and Bitcoin to demonstrate the formatting standards.

Google renamed its Dataplex Universal Catalog to Knowledge Catalog on June 25, 2026. The updated service natively ingests OKF bundles. This integration outputs a context graph designed to ground agents in verified enterprise data and reduce hallucination rates. Standardizing your context format is a critical first step when you need to evaluate and test AI agents on domain-specific tasks.

Structural Interoperability

OKF targets the context-assembly problem, where autonomous systems waste compute cycles gathering fragmented information from proprietary APIs and legacy wikis. The standard provides structural interoperability for file layouts. A data warehouse can export knowledge that a standalone agent can immediately read.

The specification remains minimally opinionated. It dictates the container shape but leaves the semantic interpretation of the text entirely to the consumer. Google explicitly noted that OKF is designed for internal organizational knowledge rather than public search ranking or SEO signaling.

If you build tools that supply context to multi-agent systems, adopting the OKF layout ensures your output is usable by downstream frameworks. Standardizing on Markdown and YAML frontmatter removes the need for custom parsing logic, letting you focus on the accuracy of the organizational knowledge itself.