Google Cloud has released the Open Knowledge Format, an open specification for the context and curated knowledge that AI agents need to work accurately.
It stores organisational knowledge as a directory of markdown files with YAML metadata. The format is vendor-neutral and needs no software development kit, runtime or proprietary account.
For information managers and governance teams, the standard tackles a persistent problem. Critical knowledge sits scattered across catalogues, wikis, shared drives and code comments. That fragmentation forces every AI agent to reassemble context from incompatible sources.
Google Cloud engineers Sam McVeety and Amir Hormati pitch OKF as a “vendor-neutral, agent- and human-friendly standard for representing the metadata, context, and curated knowledge that modern AI systems need
Each OKF concept is a single markdown file, and its file path becomes the concept's identity. Concepts can cover tables, datasets, metrics, playbooks, runbooks and APIs.
A small YAML block holds queryable fields: type, title, description, resource, tags and timestamp. Markdown links connect concepts, turning the directory into a graph of relationships. The full v0.1 specification fits on a single page.
Google built OKF on three principles: minimal opinion, producer-consumer independence, and format over platform. A bundle written by a person can be read by an AI agent without translation.
Google is publishing reference tools alongside the specification. An enrichment agent walks a BigQuery dataset and drafts a concept document for each table. A static HTML visualiser renders any bundle as an interactive graph in one self-contained file.
The specification, code repository and sample bundles are available on GitHub. https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf
Google has also updated its Knowledge Catalog to ingest OKF and serve it to the company's agents.
A Crowded Standards Field
OKF enters a busy market for AI interoperability standards. Anthropic's Model Context Protocol, donated to the Linux Foundation in December 2025, standardises how agents connect to live tools and data.
The two standards address different layers. OKF formalises portable, version-controlled knowledge, while MCP governs real-time links between agents and systems.
The release lands as enterprises rush to govern a wave of autonomous agents. Deloitte forecasts half of enterprises using generative AI will deploy agents by 2027.
Governance gaps carry measurable cost. Gartner projected 60 per cent of AI projects lacking AI-ready data would be abandoned through 2026.
Portable knowledge holds clear appeal for records and compliance managers. OKF bundles live in version control, carry timestamps and stay readable without proprietary tooling. Those traits support auditability and lineage, two priorities for governance, risk and compliance teams.
The format builds on the LLM-wiki pattern popularised by AI researcher Andrej Karpathy. He wrote that large language models “don’t get bored, don’t forget to update a cross-reference.”
Google frames OKF as a starting point, not a finished standard. “The format itself is the contribution,” the authors write. They invite producers, consumers and extensions from beyond Google's own products.