Build an OKF brain like mine!

Standardizing knowledge for the future of AI agents

My last piece on Google’s Open Knowledge Format (OKF) was one of the most popular I’ve ever published. Since then, I have been heads down working on my own OKF structure. I’ve essentially built my own personal brain, and today I want to show you how it works and why this standard is so incredibly important for the shift toward an agentic web.

The beauty of OKF is that it follows a very simple, standardized structure. While many people have pointed out that markdown files are nothing new, what Google has done is create a standard. This means if I give an agent my OKF files, it knows exactly how to read them without needing custom software. It is a universal language for AI agents.

Watch the video of me showing my OKF Brain

Understanding the YAML frontmatter, Index and Markdown files

Every OKF file starts with what is called YAML frontmatter. This is a small block of metadata at the top of your markdown file that tells an agent exactly what it’s looking at. In my personal brain, I use specific types for everything. I have concepts, entities, playbooks, references, and systems.

Here is what my folder structure looks like for my OKF brain:

(I am not sure why Antigravity misspelled "entities" in my brain. 😂)

And here is the markdown code for one of the concepts in my brain:

When an agent views my OKF, it first views the index file. This index.md file is essentially an index of the different areas the agent can access in my brain. This way, instead of my agent having to do RAG across everything in my knowledge base, it can focus specifically on the areas that are relevant.

The system I built also works to connect related concepts. When it ingests a new piece of content, whether it's a blog post I've written, a Google announcement or a piece of research, it finds what other concepts we have created that should be connected. The result is a really cool graph of all of the knowledge we have in the brain. This idea of agents extracting concepts and making connections comes from Andrej Karpathy's LLM Wiki idea.

Visualizing your knowledge graph

When you use OKF correctly, you can visualize your brain as a connected graph. Every dot in my system is a markdown file. You can see how my concepts for AI Overviews connect to my references from Google’s documentation and my internal playbooks. It’s a living, breathing map of everything I know about SEO and AI.

I’m also automating how I ingest information. I have a system that checks Google’s documentation daily. If they update something, like the documentation on AI Answers or search console features, my brain notifies me and automatically updates the relevant reference files. I don’t have to rely on my biological hardware to remember every small change anymore. My OKF brain has access to far more than I could ever keep in my head at once.

I can then query my brain and it will write something for me that is synthesized from my notes and the other information I've chosen to store in my brain. If something seems off to me, then I can simply ask the agent within my brain to correct it. Also, my agent is always looking for ways to improve and connect the information that is in the brain.

Building playbooks to save days of work

One of the most exciting things I’ve built in my OKF brain is a set of playbooks. For example, I have a playbook for generating client proposals. Usually, this is a tedious process that takes a lot of manual work. Now, my agent (Antigravity) follows the steps laid out in my proposal playbook to draft everything using my specific voice and logic.

I also built a playbook for analyzing site impact after a Google update. I created a procedural checkpoint for my agent to follow when analyzing these shifts. What used to take me two days to analyze now takes a matter of hours, and the report produced is something I’m incredibly happy with. This is the power of documenting your wisdom in a way that agents can actually execute.

I really encourage you to start experimenting with this. You don't need to be a coder to start. You can tell an agent to help you build your first OKF bundle based on the documentation Google has provided. It’s making me more productive, and I think it’s the key to staying relevant as Search continues to evolve.

To build something similar, give these links to your agent (Claude Code / Cowork, ChatGPT Codex or my favourite, Google's Antigravity) and try this prompt:

"I want to build an OKF system similar to Marie's. Read these links and then give me some ideas of what this would look like. Then, ask me questions one at a time so that together, we can decide what we want to build:

https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing/

https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md

https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

https://www.mariehaynes.com/okf/ https://youtu.be/esYAIA6lU-s"

Or, join us in the Search Bar for up to the minute news on Search and AI. Or better yet, join my paid community where we have regular calls to brainstorm on using AI and I share tips and ideas as well.