OKF v0.1 Fits on a Single Page, But Kills the Agent Context Fragmentation Mess

OKF v0.1 specification fits on a single page, but it's designed to replace the tangled mess of vendor-specific knowledge catalogs, wiki vaults, and bespoke agent contexts that plague every organization building AI agents. Andrej Karpathy's LLM wiki gist captured the pattern — give agents a shared markdown library that grows more useful over time — but every team implemented it in isolation, with no interoperability. Google Cloud's Open Knowledge Format formalizes the conventions needed to make those patterns portable across systems, agents, and organizations.

Why internal knowledge is stuck in vendor lock-in

Most organizations keep their internal knowledge — table schemas, metric definitions, incident runbooks, API deprecation notices — scattered across metadata catalogs with proprietary APIs, wikis in shared drives, code comments, and the heads of a few senior engineers. When an AI agent asks "How do I compute weekly active users from our event stream?" it has to assemble the answer from these mutually incompatible surfaces. Every vendor offers its own catalog and SDK; none of the knowledge is easily portable. The result: every agent builder solves the same context-assembly problem from scratch, while each catalog vendor reinvents the same data models.

OKF's design: markdown files with YAML frontmatter, nothing more

An OKF bundle is a directory of markdown files, each representing one concept (table, metric, playbook, API). The file path is the concept's identity. Each document carries a small block of YAML frontmatter for structured fields — type, title, description, resource, tags, timestamp — and a markdown body for everything else. Concepts link to each other with standard markdown links, turning the directory into a graph. Bundles can include optional index.md and log.md files. The spec requires exactly one thing: a type field. Everything else is left to the producer. No complex compression, no new runtime, no required SDK. A human can hand-author it; an LLM can generate it; any search tool can index it.

Reference implementations prove it works — three sample bundles shipped

To make the format concrete, Google Cloud shipped: an enrichment agent that walks a BigQuery dataset and drafts OKF concept documents for every table and view, then runs a second LLM pass to enrich with citations and join paths; a static HTML visualizer that turns any OKF bundle into an interactive graph view with no backend; and three ready-to-browse sample bundles — GA4 e-commerce, Stack Overflow, and Bitcoin public datasets (actual URLs in the source repo). These are proofs of concept, deliberately. The agent demonstrates one way to produce OKF; the visualizer shows one way to consume it. Nothing about the format requires a specific framework or LLM.

OKF v0.1 is a starting point. Google Cloud has also updated its Knowledge Catalog to ingest OKF. The spec lives on GitHub, versioned and designed for backward-compatible growth. If the ecosystem adopts it, OKF could become the lingua franca for knowledge exchange — not another platform, just a format that any party can speak.

Source: Introducing the Open Knowledge Format
Domain: cloud.google.com