The Complete Guide to Open Knowledge Format (OKF)
Everything you need to know about OKF — from what it is and why it matters, to creating your own bundles, choosing the right tools, and discovering bundles in the ecosystem. Updated for July 2026.
1. What is the Open Knowledge Format (OKF)?
Open Knowledge Format (OKF) is a vendor-neutral, filesystem-based standard for packaging knowledge as plain markdown files. Launched by Google Cloud Platform in June 2026, OKF is designed to be authored by humans, consumed by AI agents, and managed in version control.
Key Characteristics
- Filesystem-native: No SDKs, databases, or APIs required. A bundle is just a directory with markdown files.
- YAML frontmatter: Each file has structured metadata (title, description, type, tags, relationships) between
---delimiters. - Git-versioned: Bundles live in git repositories, enabling collaboration, review, and version history.
- Agent-native: AI agents like Claude Code, Cursor, and OpenCode can load and query bundles directly.
- Vendor-neutral: No platform lock-in. OKF bundles work with any tool that reads markdown.
Why Google Created OKF
AI agents need structured, reliable knowledge to function effectively. Traditional approaches like vector databases require complex infrastructure, while ad-hoc approaches (prompts, CLAUDE.md files) don't scale. OKF fills the gap: a standard format that agents can discover, load, and reason over without custom integration.
2. Why OKF Matters for AI Agents
Before OKF, every AI agent workflow had to reinvent knowledge management. The options were:
- System prompts: Static, limited context window, hard to update
- RAG pipelines: Requires vector DB, embedding models, infrastructure
- CLAUDE.md / .cursorrules: Single files, no structure, no relationships
- Custom formats: Every tool invented its own approach
OKF solves this by providing a universal standard that any agent can consume. The ecosystem has grown from zero to 288 bundles by 254 authors in just three weeks — a signal that the format is filling a real need.
3. OKF vs Other Knowledge Formats
| Feature | OKF | llms.txt | MCP | CLAUDE.md | Vector DB |
|---|---|---|---|---|---|
| Format | Multi-file directory | Single file | Protocol | Single file | Database |
| Knowledge scope | Full domain | Index/summary | Tool calling | Project rules | Unstructured |
| Agent-native | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes | ⚠️ Needs pipeline |
| Version control | ✅ Git-native | ✅ Git-native | ❌ Server-based | ✅ Git-native | ❌ |
| Structured metadata | ✅ YAML frontmatter | ❌ Plain text | ❌ | ❌ Plain text | ⚠️ Embeddings only |
| Discovery | ✅ BundleDex | ✅ llmstxt.org | ✅ MCP directories | ❌ Manual | ❌ |
| Best for | Knowledge packages | Site summaries | Real-time tools | Project rules | Search |
These formats are complementary, not competitive. A well-architected AI agent setup uses all four: llms.txt for discovery, OKF bundles for knowledge, MCP for tool access, and CLAUDE.md for project-specific rules. See our detailed comparison →
4. How to Create an OKF Bundle
Creating an OKF bundle is straightforward. Here's the step-by-step process:
Create the directory structure
my-knowledge-bundle/
index.md
SPEC.md
concepts/
concept-1.md
concept-2.md Write index.md with YAML frontmatter
---
name: my-knowledge-bundle
description: A bundle of knowledge about [topic]
version: 0.1.0
okf_version: 0.1.0
--- Add concept files
Each concept file represents a discrete unit of knowledge. Use YAML frontmatter to describe metadata:
---
title: Concept Name
description: What this concept covers
type: concept
tags: [tag1, tag2]
--- Push to GitHub and add the OKF topic
Create a GitHub repository, push your bundle, and add the okf topic. BundleDex's automated discovery will find and index it within 24 hours.
Get listed on BundleDex
Once indexed, your bundle appears on bundledex.net. You can also manually submit at /submit. Add the BundleDex badge to your README to show your bundle is listed.
5. OKF Tools & Ecosystem
The OKF ecosystem includes tools for creating, validating, and consuming bundles:
BundleDex
Directory of 288 OKF bundles. Search, discover, and use knowledge packages for AI agents.
bundledex.netOKFy
Automatically convert existing documentation into OKF format. Supports markdown, HTML, and more.
Open Knowledge CLI
Command-line tool for validating and managing OKF bundles. Check conformance, list concepts, and more.
okf-knowledge
A Claude Code skill (/okf) to create, read, maintain, and visualize OKF bundles directly from the editor.
OKF Conformance Checker
Validates bundles against the OKF specification. Ensures your bundle passes the conformance test.
BundleDex MCP Server
Agent-native discovery via MCP protocol. Query bundles from any MCP-compatible agent.
/api/mcp6. Best OKF Bundles to Use
The top bundles by community adoption show the range of what OKF can do:
Markdown memory system for you and your AI agent
Distill videos, PDFs, transcripts, and notes into source-backed Agent Skills. Uses OKF/OKF-format for structured knowled
Persistent memory plugin for OpenCode. Obsidian-style knowledge base that survives across sessions
OKF-powered knowledge context for Claude Code — injects your project's knowledge base at every session. Includes OKF-con
AI chat, workflow automation, semantic search (RAG), LLM Wiki (OKF), and dashboards with reading memos powered by Google
7. How to Find OKF Bundles
There are several ways to discover OKF bundles:
Via BundleDex (Recommended)
- Web interface: Browse and search at bundledex.net
- JSON API:
bundledex.net/api/bundles.json— returns all bundles as structured data - MCP Server:
bundledex.net/api/mcp— query from any MCP-compatible agent - llms.txt:
bundledex.net/llms.txt— LLM-friendly summary of all bundles
Via GitHub
- Search GitHub for repos with the
okftopic - Browse the
google/okfofficial repository - Check
awesome-okffor curated collections
Via Hugging Face
- The BundleDex dataset is available at
Rex-McClawd/bundledex-okf-bundles - Contains all indexed bundles with metadata for programmatic access
8. Frequently Asked Questions
Is OKF free to use?
Yes. OKF is an open standard with no licensing fees. All bundles are open source under their respective licenses.
Do I need Google Cloud to use OKF?
No. OKF is vendor-neutral. While Google created the spec, bundles work with any tool that reads markdown files.
Can OKF bundles include images or binary files?
OKF is designed for text-based knowledge. Images can be referenced via URLs, but the bundle itself is markdown and YAML.
How is OKF versioned?
OKF uses semantic versioning. The current spec version is 0.1.0. Each bundle declares its okf_version in index.md.
Can I use OKF with any AI agent?
Any agent that can read markdown files can use OKF bundles. Claude Code, Cursor, and OpenCode have native support. For other agents, you can load bundles via the JSON API or MCP server.
Ready to explore OKF?
Browse 288 OKF bundles on BundleDex, or create your own and submit it to the directory.
Browse Bundles → Submit a Bundle →