What Is OKF (Open Knowledge Format)? Google’s Open Standard, Explained

Open Knowledge Format (OKF) is an open, vendor-neutral standard for packaging information as graph-linked Markdown files — with YAML frontmatter — that AI agents can read natively. If you have searched for "OKF format" or "Google’s OKF" and found mostly jargon, here is the plain-English version.

The problem OKF solves

Most organizational knowledge lives in places an AI agent cannot use well: PDFs, dashboards, bespoke databases, and APIs that each need custom integration. To give a model real context you normally dump documents into a vector store and hope retrieval surfaces the right chunk. That loses structure — the relationships between things — and it is hard to verify what the model actually saw.

What OKF actually is

OKF takes a different approach. Information is written as plain Markdown files, one per entity, each with YAML frontmatter for typed metadata and explicit links to related files. The result is a knowledge graph an agent can traverse — follow a link from a requirement to the control that satisfies it, to the asset it protects, to the test that proves it. It is human-readable and machine-readable at the same time.

• Plain Markdown — readable by people, diffable in Git, openable in any editor
• YAML frontmatter — typed metadata (IDs, status, dates, relationships)
• Graph links — entities reference each other, so an agent can walk the graph
• Open and vendor-neutral — no proprietary database or API required to read it

Why “Google’s OKF”?

OKF originates as an open standard backed by Google for exactly this need — a portable way to hand AI agents structured context without lock-in. Because the format is open, any tool can produce or consume it; you are not tied to the vendor that generated the bundle.

OKF in practice: a compliance graph

A concrete example: CATAAM’s OKF Context Engine serialises an entire compliance program — requirements, controls, assets, automated tests, and security findings — into OKF. Each bundle is Ed25519-signed, so it is a verifiable point-in-time record, and you can sync it to your own Git repo or download it on demand. An assistant like Claude can then answer "which controls cover this S3 bucket?" from your real, current data. The same graph doubles as evidence for ISO 42001 AI governance — proving which AI systems are in scope, tested, and monitored, not just claimed.

Where to go next

If you are building with agents, read OKF for AI agents. If you care about delivery and version control, see OKF and Git. And to understand how OKF and the Model Context Protocol fit together, read OKF vs MCP.

Frequently asked questions

What is Open Knowledge Format (OKF)?

OKF is an open, vendor-neutral standard for packaging information as graph-linked Markdown files with YAML frontmatter that AI agents can read natively. It keeps the relationships between entities intact, so an agent can traverse the knowledge graph rather than relying on document chunks.

Is OKF a Google standard?

OKF (Open Knowledge Format) is an open standard backed by Google for giving AI agents portable, structured context without vendor lock-in. Because it is open, any tool can produce or consume OKF bundles.

How is OKF different from a vector database?

A vector database stores unstructured chunks for similarity search and loses the explicit relationships between items. OKF preserves structure as a linked graph of Markdown files, so an agent can follow defined links between entities and you can verify exactly what it read.