OKF gives every agent durable company knowledge from a folder of typed markdown

OKF, the Open Knowledge Format from Google Cloud, is a folder of plain markdown files in Git where each file describes one thing (a table, a metric, a runbook) and carries a single required field: type. An agent opens the file and reads it like a new hire reading the handbook, with no retrieval pipeline and no vector database. The files link with ordinary markdown links into a navigable graph, and each folder can hold an index file the agent reads first to decide what to open, a pattern the spec calls progressive disclosure.

The format is deliberately forgiving: a consumer must tolerate a broken link, an unknown field, or an unseen type, because real knowledge is always half-finished. The analogy is precise: if MCP (see: MCP is a protocol layer that decouples tools from providers) is the socket connecting an agent to tools and live data, OKF is the knowledge flowing through it. It is curated facts an agent navigates on purpose, not fuzzy search over hopeful chunks (see: an LLM wiki compounds where RAG rediscovers), and it is durable memory outside the model window (see: agent memory persists context that the model window cannot hold) that belongs to the harness (see: memory is part of the harness not a separate tool).

Because it is just files in version control (see: git-native workflows make every change a checkpoint), the same documents serve humans and agents at once. OKF fits stable expertise, definitions and schemas and how-it-really-works; for data that changes faster than once a month, keep it behind an API, not a document.