Historical source material recovered on 2026-09-09. Capture dates and source limitations apply; recovery is not fresh verification.

Open Knowledge Format

Open Knowledge Format, or OKF, is Google Cloud's proposed open specification for representing AI-usable knowledge as markdown files with YAML frontmatter and shared conventions.

Key points

• Google Cloud describes OKF v0.1 as a vendor-neutral, agent- and human-friendly standard for metadata, context, and curated knowledge ^[src-100].
• The format formalizes the LLM Wiki Pattern: directories of markdown files, YAML frontmatter, readable content, and conventions that agents can consume without translation ^[src-100].
• OKF is strategically important because it treats enterprise context as portable infrastructure rather than a proprietary memory silo ^[src-100].
• It also validates Robin's knowledge graph design choices: markdown, frontmatter, source IDs, explicit relationships, and human-readable pages can serve both people and agents ^[src-100].

Related entities

• Google Cloud
• Google Cloud Blog

Related concepts

• LLM Wiki Pattern
• Context Engineering
• Enterprise Knowledge Graph
• Agentic Data Cloud
• Machine Scannable Content

Source references

• ^[src-100] Sam McVeety and Amir Hormati – "How the Open Knowledge Format can improve data sharing" (2026-06-12)

Historical recovered snapshot — 2026-08-21

Preserved from ai-watch-src377-okf-repository-20260821-065755. Documentary recovery on 2026-09-09; claims below retain their historical source scope and were not freshly verified. Current synthesis above remains authoritative.

Open Knowledge Format, or OKF, is Google Cloud's proposed open specification for representing portable knowledge as directory trees of human-readable Markdown with YAML frontmatter and shared conventions.

• OKF v0.2 defines minimal bundle conformance: parseable frontmatter and a non-empty type on non-reserved Markdown files, with additional conventions for index.md and log.md ^[src-377].
• Consumers should tolerate unknown fields and types, missing optional fields, broken links and absent indexes. This makes the format extensible without requiring every participant to share one fixed ontology ^[src-377].
• Bundle-root links, progressive-disclosure indexes and append-style logs make the files navigable by both people and agents ^{[src-100] [src-377]}.

• Provenance belongs in structured sources, while generated and verified distinguish production evidence from review evidence ^{[src-281] [src-377]}.
• Derived trust tiers — unverified, machine-confirmed and human-reviewed — should be treated as evidence states, not guarantees that a claim is true ^[src-377].
• Lifecycle fields such as status and stale_after make freshness explicit and machine-actionable ^{[src-281] [src-377]}.
• Attested Computation records runtime, parameters, computation, executor and attester. An attestation confirms a declared execution path; it does not by itself establish semantic correctness ^[src-377].

• Google's Apache-2.0 subtree includes a reference producer, browser graph visualiser and four sample bundles. At inspected commit 62432a095456147ee71e70ac6e4dc0d2dea3ac30, its own test suite passed 39/39 ^[src-377].
• The producer and visualiser are explicitly proofs of concept. Receipt and verdict wire formats, attester ABI and sandboxing, caching and semantic-layer templates remain deferred ^[src-377].
• Version 0.2 includes breaking migrations from timestamp to generated.at and from body citations to frontmatter sources, despite being labelled a minor version ^[src-377].

• OKF treats enterprise context as portable infrastructure rather than a proprietary memory silo ^{[src-100] [src-377]}.
• It supports the LLM Wiki Pattern: files remain readable and versionable, while search, graph and agent consumers can be derived from the same source ^{[src-100] [src-377]}.
• AI Watch already aligns through Markdown, frontmatter, source IDs, current-truth metadata and logs. The repository adds a useful distinction between creating a record, verifying it and attesting a specific computation ^[src-377].
• Broad interoperability and production adoption are not yet demonstrated by the repository itself ^[src-377].

• ^[src-100] Sam McVeety and Amir Hormati — “How the Open Knowledge Format can improve data sharing” (2026-06-12)
• ^[src-281] Google Cloud — “Open Knowledge format v0.2 tackles agentic trust” (2026-07-24)
• ^[src-377] GoogleCloudPlatform — “Open Knowledge Format v0.2 specification and reference implementation” (repository snapshot inspected 2026-08-21)