Plain-English explainer
A Knowledge Bundle is a structured collection of useful information, sources, citations, metadata, and guidance that an AI assistant or agent can inspect and reuse.
A strong bundle makes the knowledge clear to both people and tools. It should show where information came from, what it can be used for, and how it should be handled.
Most AI workflows lose time rebuilding context. Bundles make that context portable, inspectable, and easier to improve over time.
A slower explanation
A Knowledge Bundle is like a well-organised project folder. It does not just contain notes. It explains what the notes are about, where facts came from, how the pieces connect, and what the AI should look at first.
Google's OKF spec calls a Knowledge Bundle the unit of distribution: a self-contained collection of knowledge documents. In simpler words, it is the thing you can share, download, inspect, and give to an assistant.
A concept is one markdown file inside the bundle. It might explain a metric, an API, a table, a topic, a workflow, or a playbook. The file path becomes the concept's identity.
At the top of each concept is a small YAML block. The required field is type. Recommended labels include title, description, resource, tags, and timestamp.
After the label comes normal markdown: headings, lists, examples, tables, and notes. This is where the useful knowledge lives in a form humans can read and agents can parse.
Tiny example
The label gives the AI a quick preview. The body gives the actual instructions, examples, links, and source notes.
---
type: Playbook
title: New customer onboarding
description: Steps for preparing a customer success handover.
tags: [customers, onboarding]
---
# Steps
1. Confirm the customer plan.
2. Read the handover notes.
3. Check the setup checklist.
# Citations
[1] Customer onboarding policyConcepts can link to other concepts with ordinary markdown links. That means the bundle can behave like a small knowledge map, where an assistant can move from a metric to a table, from a table to a playbook, or from a playbook to a source document.
An index.md file is like a table of contents. It helps a person or agent see what is in a folder before opening every file.
A log.md file can record updates over time. That is useful when knowledge changes and buyers want to understand whether a bundle is being maintained.
When a concept makes claims from external material, the OKF spec recommends a citations section. This helps keep knowledge grounded instead of leaving the AI to guess.
Google source
Google Cloud introduced Open Knowledge Format as an open, portable way to represent the metadata, context, and curated knowledge that AI systems need. The launch article introduced v0.1; the public spec is now v0.2 and describes a minimal structure based on markdown files, YAML frontmatter, links, provenance, trust, lifecycle, and freshness signals.
Next step