Google's Open Knowledge Format: Just Markdown for AI Agents
A portable concept-and-bundle contract for agent knowledge, with reference producer and consumer tools
Prism Labs explains Google Cloud’s v0.1 Open Knowledge Format proposal: concepts are Markdown files with minimal YAML metadata, ordinary links between them form a graph, and optional index and log files add navigation and history. Google supplies a BigQuery enrichment agent, a graph visualizer, sample bundles, and Knowledge Catalog ingestion as first-party proofs of concept. The recording shows no independent cross-vendor interchange or retrieval-quality evidence, and it misstates which fields are required.
Agents need knowledge such as table schemas, metric definitions, and outage runbooks, but that information is often scattered across incompatible tools. Prism Labs explains Google Cloud's Open Knowledge Format (OKF) v0.1 proposal: exchange that knowledge as a directory of Markdown files that people and agents can read with existing tools.
The Format Mechanism
- One file describes one concept: A concept might be a table or a metric. YAML metadata above the Markdown body makes it possible to identify and filter documents. The walkthrough shows
type,title,description,resource,tags, andtimestamp; onlytypeis required, despite the video's closing claim that all six are needed 01:52–02:48(opens in a new tab), 03:41–04:40(opens in a new tab), 06:20(opens in a new tab). - Paths identify concepts; links connect them: Under the v0.1 specification(opens in a new tab),
tables/orders.mdhas the IDtables/orders. Ordinary Markdown links connect related concepts. An optional index lets an agent see what is available before opening individual files; optional logs record changes over time 02:48–03:41(opens in a new tab). - Writers and readers share a format, not a service: A person or agent can produce a bundle independently of the tool that consumes it. This makes interoperability a design goal, though the video does not demonstrate exchange between independent vendors 03:41–04:40(opens in a new tab).
- Google supplies a starting implementation: The video describes a BigQuery enrichment agent that drafts documents and adds supporting context, a static graph visualizer, three public-data sample bundles, and Knowledge Catalog ingestion. These first-party tools illustrate how to create and consume bundles, without establishing their accuracy or operational scale 03:41–05:30(opens in a new tab).
- Plain files make inspection easy, not maintenance automatic: The repository walkthrough lists the samples and reads a concept directly in a terminal. The narrator also proposes letting agents keep the knowledge current, but does not demonstrate that they can maintain a large knowledge base accurately 04:40–06:20(opens in a new tab).