OKF of Google Transforms AI Agents’ Knowledge Into Plain Text!

Google Cloud has announced a new open standard for accessing corporate knowledge by AI agents: the Open Knowledge Format (OKF). The current iteration (version 0.1) is essentially a folder full of markdown files with yaml headers. Each file represents a ‘concept’, which has a single required attribute (type), and points to other concepts using markdown links, thus building out a knowledge graph. There’s no proprietary account or SDK, no cloud platform lock-in, and bundles can be rendered on github, distributed as tarballs, and mounted on any filesystem.

This is an attempt to solve a boring but costly problem: most organizations have knowledge about their operations spread out over a variety of disconnected systems, so each team developing an AI agent has to solve the “how do I get context?” problem for themselves. OKF aims to solve this by making this knowledge portable and interoperable.

1How it differs from RAG

You may be familiar with the concept of retrieval-augmented generation (RAG). Essentially, you throw documents into a database, then retrieve chunks of text when asked a question and paste those chunks into the prompt. OKF takes a different approach. Rather than dynamic retrieval of chunks, it provides curated, linked concepts for agents to read (and edit!). Google follows three core principles: “minimally opinionated (single required field), consumer/producer independent (humans and agents both read and write the same files), and ‘format, not platform.’ They also released some reference implementations to help people get started: a BigQuery enrichment agent, a static HTML visualization tool, and three example bundles.

2Why this matters to you

This is an interesting development even if you’re not interested in writing yaml headers yourself. It’s a quiet but significant wager on how AI systems should remember things. By storing your corporate knowledge in markdown files under version control, it becomes diffable, reviewable, and moveable — just like code. Rather than letting your knowledge languish in some untrusted notion space, you can now make it portable and interoperable. Additionally, since it’s an open standard rather than a Google product, you won’t be beholden to the whims of Google’s product managers and pricing schemes.

As someone who has spent a fair amount of time providing context to coding agents, I think the most interesting thing about this is that humans and agents read the same files. There is no “AI layer” to worry about keeping consistent; what the agent sees is what you can edit in your favorite text editor. Regardless of whether OKF succeeds or another format emerges, this idea is worth following.