原文 · 未翻译
Google Cloud's Open Knowledge Format turns scattered docs into Markdown files for AI agents
Google Cloud is introducing the Open Knowledge Format (OKF), a new spec that standardizes knowledge as Markdown files and makes it portable across systems.
It takes the "LLM wiki" pattern recently popularized by Andrej Karpathy and turns it into an interoperable format. OKF v0.1 represents knowledge as a directory of Markdown files with YAML frontmatter. The spec is minimal. One required field ("type"), a handful of optional fields like title, description, resource, tags, and timestamps, plus a Markdown body for everything else.
Concepts link to each other through standard Markdown links, forming a knowledge graph. An OKF bundle is readable in any editor, renders on GitHub, and can be indexed by any search tool.
Fragmented knowledge slows AI agents down
Most organizations know the problem OKF is trying to solve. Knowledge is scattered across metadata catalogs, wikis, code comments, notebook cells, and the heads of individual engineers. When an AI agent needs to write a SQL query for a specific dataset, it has to piece together fragments from all these sources.
According to Google Cloud, every agent developer currently solves this context problem from scratch, and every catalog vendor reinvents the same data models. Obsidian Vaults hooked up to coding agents, AGENTS.md and CLAUDE.md convention files, "metadata as code" repos on data teams. They all follow a similar pattern. But each solution is custom-built and not designed to work with the others, Google says. Knowledge stays locked inside the system that created it. That's the gap OKF aims to close.
Built to be minimal and portable
OKF requires just one field ("type"). Which types exist, what extra fields a document contains, and how the body is structured is up to the producer. Producers and consumers are decoupled. A bundle written by humans can be consumed by an AI agent. A machine-generated bundle can be viewed in a visualizer. OKF works with any cloud provider, database, or agent framework.