Quick start: OKF in five minutes

By the end of this page you will have built a valid OKF bundle, loaded it into the Hub, and asked your first question of it. No SDK, no build step. A text editor is enough.

If you just want to watch first: sign in to the Hub with Google and load the demo database. The demo does not count against your upload limit. If you bring your own knowledge, read on.

Step 1: Write one concept

A concept is a Markdown file with a small YAML header. Create a file crawl-budget.md:

---
type: Guide
title: Crawl-Budget bei großen Websites
description: Wie Google das Crawl-Budget verteilt und woran man Verschwendung erkennt.
tags: [crawl-budget, technical-seo, indexierung]
---

# Was Crawl-Budget bedeutet

Crawl-Budget ist die Zahl der URLs, die Google in einem Zeitraum
auf einer Domain abruft …

# Woran man Verschwendung erkennt

Endlose Parameter-URLs, Session-IDs und Filterkombinationen …

Only type is required. The other three fields cost you ten seconds and noticeably improve search, graph, and retrieval. The full field reference is in the frontmatter schema.

Step 2: A second concept and a link

Knowledge gets interesting the moment concepts reference each other. Create logfile-analyse.md and link the first concept bundle-relative, with a leading /:

---
type: Guide
title: Logfile-Analyse für Crawl-Steuerung
description: Aus Server-Logs ablesen, was Google wirklich crawlt.
tags: [logfiles, crawl-budget, technical-seo]
---

# Ziel

Server-Logs zeigen das echte Crawl-Verhalten, nicht das vermutete.
Das ergänzt die Sicht aus [Crawl-Budget](/crawl-budget.md).

Both files share the tags crawl-budget and technical-seo. Those shared tags later create the similarity edge in the Knowledge Graph; the link creates the directed edge.

Step 3: Package as a bundle

A bundle is simply a folder. Put both files in it and zip it:

mein-wissen/
├── crawl-budget.md
└── logfile-analyse.md

zip -r mein-wissen.zip mein-wissen/

That was the entire "format." Optionally add an index.md as an overview and a log.md as a change history – both are reserved filenames with a fixed structure, described in the specification.

Step 4: Load into the Hub

Sign in to the Hub with Google, click "Upload OKF," and choose "Ready-made OKF bundle." Allowed formats are .zip, .md, and .json, up to 20 MB per file. Upload mein-wissen.zip.

The Hub reads the Markdown files, creates two concepts, and builds a semantic search index. System files and the aggregate files README.md, INDEX.md, okf.md are skipped on purpose.

Step 5: Ask and check sources

Open the AI RAG Assistant, select your new source and a model, and ask a focused question: "How do I spot wasted crawl budget, and how does logfile analysis help?" The assistant retrieves the relevant concepts, answers only from them, and numbers the sources it used. Selecting a source number opens the corresponding OKF excerpt in the reader.

Check the sources before using an answer. RAG reduces unsupported statements but does not fully prevent them.

What helps next

Three small habits make every further bundle better. Write descriptions that stand on their own – search weights title and description above the body. Keep tags consistent, or the similarity edges splinter. And split large topics into several concepts instead of inflating one concept to 5000 words; smaller, linked units retrieve more precisely.

Continue

Already have knowledge in Notion, Obsidian, Logseq, or Confluence? The migrations page turns exports straight into OKF. To generate and check bundles automatically, see CLI and CI/CD. Ready-made enrichment prompts are under prompt templates.