Open Knowledge Format: Google standardizes AI agent memory

Google just published the Open Knowledge Format (OKF), an open spec for representing knowledge — table schemas, incident runbooks, metric definitions, API docs — as a directory of Markdown files with YAML frontmatter. No SDK, no central registry. If you can git clone a repo, you can consume an OKF bundle.

In most organizations, the context AI agents need is scattered: proprietary catalog APIs, Confluence wikis, docstrings, a few engineers who hold it in their heads. Every team ends up reinventing the same scraping logic to pull it together, and the result stays locked behind whatever tool produced it.

OKF formalizes what many teams were already doing informally — the AGENTS.md pattern, Obsidian vaults wired to coding agents, Karpathy’s LLM wiki. What it adds: just enough convention so that different producers and consumers can talk to each other without a translation layer. One required field: type. Everything else is optional.

The spec is genuinely vendor-neutral, and Google keeps it clearly separate from their own tooling (kcmd, Knowledge Catalog, Dataplex). You can adopt OKF without touching GCP. v0.1 ships with a reference BigQuery enrichment agent and a static visualizer.