OKF tool: build, read and query your own Open Knowledge Format

The Open Knowledge Format (OKF) is Google’s open standard, live since June 2026, for knowledge that humans and AI agents read alike. Technically: a directory of Markdown files with YAML frontmatter whose concepts link into a graph.

Google published the standard and invited third parties to build their own tools for it. But the building blocks it ships with target data teams on BigQuery, not ordinary users.

That is exactly where my lab experiment comes in. The tool makes OKF usable without you writing Markdown and YAML by hand. Upload once, then structure, see it as a graph and query it.

Create and structure – Concept Library

You upload a bundle as .zip, .md or .json in the OKF v0.2 format. The parser reads the concepts along with their frontmatter — titles, descriptions, tags and body text. Each concept stays a file, exactly as the spec intends, and you edit it in place.

See and understand – knowledge graph

The concepts link to each other through ordinary Markdown links, which turns your bundle into a graph. A force-directed view shows how topics connect and where the links run. It also shows the gaps — the links you would expect but do not yet have. A click on a node opens the concept behind it.

Ask and answer – RAG chat

You ask in plain language across your entire OKF. Gemini 2.5 Flash answers exclusively from your bundle, so it stays within what you uploaded. Every statement points back to the concept it came from. That way you can check each line against its source.

Why OKF

Google released OKF as a format, not a platform. It stays vendor-neutral, lives in version control next to your code and needs no SDK to read or write. The same files serve a human in the editor and an agent reasoning over the bundle.

Anyone already working with llms.txt, Obsidian or AGENTS.md and CLAUDE.md knows the pattern. OKF is the shared form those patterns were missing. If you work on context engineering, this is where you build the knowledge base an agent draws its context from.

The spec to read up on: Open Knowledge Format on GitHub.

Who it is for

For developers who want to give their AI agent portable context. For data and research teams whose knowledge is stuck in a single vendor’s catalog. And for anyone who wants a knowledge base a model reads without guessing.

Status and limits

OKF itself is at version 0.2; the standard is still evolving. My tool is an experiment and grows with it. The chat currently runs on Gemini 2.5 Flash, so answer quality stands and falls with the model and with the cleanliness of your uploads. I build this for my own use and open it up step by step. Feedback is welcome.

Contact

Want to try the OKF tool or have an idea for it? Write to me.