Google Cloud just released a new way to feed AI agents the company info they need. It is called the Open Knowledge Format. It turns scattered docs into simple Markdown files that any agent can read. The goal is to stop every AI builder from reinventing the wheel.
The Gist
- The Open Knowledge Format saves company knowledge as Markdown files with simple tags
- Every AI agent can read the same files, no matter which platform built it
- Google released sample bundles and a BigQuery enrichment agent on GitHub
What the Open Knowledge Format actually is
Google Cloud rolled out a new specification called the Open Knowledge Format, often shortened to OKF. The idea is simple. Instead of letting every company store its knowledge in a different system, OKF defines one shared file format. Any AI agent can read it.
The format is built on Markdown. Markdown is a lightweight way to write text with simple styling, the kind you see in GitHub READMEs or in chat tools like Slack. Each file starts with a small block of tags called YAML frontmatter, which describes what the file is about.
The Open Knowledge Format only requires one field to start. That field is “type”. Everything else, like the title, description, tags, or timestamps, is optional. You can build a tiny bundle in five minutes or a giant one across thousands of files. The format scales without changing shape.
Files connect to each other through normal Markdown links. That creates what is called a knowledge graph. A knowledge graph is a web of concepts where each idea points to the others it relates to. Agents follow these links to gather the context they need before answering a question or taking an action.
The mess this is trying to clean up
In most companies, useful information sits in many different places. Some lives in wikis. Some in metadata catalogs that describe databases. Some in code comments. Some inside notebooks. Some only in the heads of three senior employees. Each tool stores it in its own format.
AI agents struggle with this. An agent is a piece of AI software that takes actions on its own, like booking a meeting, querying a database, or producing a report. To work well, an agent needs context. It needs to know what a customer table contains, how to read the company glossary, or which projects are confidential.
Per the Open Knowledge Format announcement, every agent developer currently solves this context problem from scratch, and every catalog vendor reinvents the same data models. Each team ends up writing custom code to plug their AI into their own knowledge base. That work is expensive, fragile, and almost never reusable.
The Open Knowledge Format aims to break that cycle. By turning everything into the same kind of Markdown file, Google gives builders one common surface. The same bundle that feeds a Google Cloud agent can feed an agent built on Anthropic, on OpenAI, or on a local model. The format does not care about the brand.
The design is borrowed from a pattern popularized by Andrej Karpathy, an AI researcher previously at OpenAI and Tesla. He called it the “LLM wiki”, a way to organize knowledge for large language models the same way Wikipedia organizes it for humans. Google turned that idea into a real open standard.
Keep learning on AI Noobies:
- KPMG AI report pulled after invented client cases
- Count Anything: the AI that counts anything you point at
- Google sues “Outsider Enterprise”, a $1.9B AI scam ring
What this changes for daily AI users
In the short term, the Open Knowledge Format is a builder’s tool. If you are not coding agents yourself, you will not interact with it directly. But the indirect effects will reach you faster than you think. The AI tools you already use are going to get more useful as they ingest these bundles.
Google released several pieces alongside the spec. An enrichment agent that converts BigQuery datasets into OKF bundles. A simple visualizer that shows your bundle as a clickable HTML page. Sample bundles based on GA4 e-commerce data, Stack Overflow content, and Bitcoin datasets. The code is on GitHub for anyone to use.
In the medium term, expect the agents you use at work to know your company better. Your sales assistant will understand your product catalog. Your support chatbot will read your internal glossary. Your data agent will pick the right table without asking you to specify the schema every time. That is what proper context unlocks.
The Open Knowledge Format also helps with portability. If your company switches AI vendors next year, your knowledge does not get trapped in a proprietary format. The same Markdown bundle keeps working. That is a quiet but important win for companies tired of being locked into a single AI provider.
For everyday users, the takeaway is simple. AI agents are about to get less stupid about your specific world. The shift will not be a flashy product launch with a new app to install. It will arrive quietly inside the tools you already use, as those tools start reading the same kind of bundles Google just standardized. The next time your AI assistant answers a question with the right context, this is part of why.
Follow the story on AI Noobies.
About the Author
