Open Knowledge Format (OKF)

An open, vendor-neutral specification from Google Cloud for representing knowledge as a directory of markdown files with YAML frontmatter. It formalizes the…

​

Open Knowledge Format (OKF)

Version drift between sources. The launch blog post describes v0.1 (only type required; fields type, title, description, resource, tags, timestamp). The okf README describes v0.2, which adds trust, provenance and freshness fields. The README is the newer source and says the spec moved to its own repository. Field details below are from these two captures only; the v0.2 SPEC.md itself was not ingested. Verify against the canonical spec before adopting any field.

An open, vendor-neutral specification from Google Cloud for representing knowledge as a directory of markdown files with YAML frontmatter. It formalizes the LLM-wiki pattern (Compounding Knowledge Base) so bundles written by one producer can be read by any consumer without translation.

• Canonical repository: GoogleCloudPlatform/open-knowledge-format. The older knowledge-catalog/okf/ copy is frozen.
• Status: early (v0.1 launch, v0.2 current per README). Explicitly designed for backward-compatible growth.

​

Core model

Term	Meaning
Bundle	A directory tree of OKF documents; shippable as a tarball or git repo
Concept	One markdown file; anything worth knowing (table, metric, runbook, API)
Identity	The concept’s file path within the bundle
Links	Ordinary markdown links between concepts, forming a graph
index.md	Optional per-directory listing for progressive disclosure
log.md	Optional chronological change history

​

Frontmatter by version

Field	v0.1	v0.2	Purpose
type	required	required	Concept kind; producer-defined vocabulary
title, description, resource, tags	yes	yes	Display and query fields; resource links the real asset
timestamp	yes	not listed in README	Last-change time
sources	—	yes	Provenance, with per-source credibility signals
generated	—	yes	Who or what produced the concept
verified	—	yes	Who confirmed it; consumers derive a trust tier from generated + verified
status	—	yes	Whether the concept is current
stale_after	—	yes	Freshness deadline

“Not listed” means only that the README’s summary of v0.2 does not mention it, not that it was removed.

​

Principles

Minimally opinionated (only type required), producer/consumer independence (the format is the contract), and format-not-platform (no account or SDK ever required).

​

Reference tooling (proofs of concept)

• Enrichment agent: a BigQuery metadata pass, then an LLM web pass that crawls seed URLs and enriches concepts, with page caps and host allow-lists enforced inside the fetch tool. See Okf V02 Readme.
• Visualizer: a single self-contained HTML graph view with backlinks, search and type filter.
• Sample bundles: GA4, Stack Overflow, Bitcoin, Acme Retail.
• Google Cloud Knowledge Catalog ingests OKF and serves it to Google agents.

​

Relevance to this wiki

This wiki is already close to OKF: markdown + YAML, one page per concept, index.md, log.md, cross-links. The gaps are mapped in OKF v0.2 vs this wiki’s frontmatter. The v0.2 trust and freshness fields are the most useful idea to borrow, because they turn “is this page stale or unverified?” into a query instead of a judgment call during lint.

​

Related

• Okf V01 Announcement · Okf V02 Readme
• Llm Wiki Pattern — the Karpathy gist OKF formalizes.
• AGENTS.md (format) — a sibling convention-file standard.
• qmd — would index an OKF bundle unchanged, since it is plain markdown.

​

Linked pages

• AGENTS.md (format)
• Compounding Knowledge Base
• OKF v0.2 vs this wiki’s frontmatter
• qmd