How to Create an OKF Bundle from a Website

Short answer

Use the LLM Wiki & OKF Bundle Generator. Give it a website (or an Apify dataset) and it writes a conformant OKF v0.2 bundle: one concept file per page with type, title, description, resource, tags, sources, generated and stale_after, an index.md per folder with okf_version at the root, a log.md, a zip and a graph viewer.

A real use case

A data team is trying Google Cloud Knowledge Catalog, which ingests OKF. Their product docs live on a public site, not in BigQuery. They run the actor on the docs, check the bundle in the viewer, and load it next to the table bundles the reference agent made, so agents see the docs and the data in one catalog.

Step by step

Open the actor on Apify and sign in.
Add start URLs, and optionally Apify datasets to document as schema concepts.
Choose AI enrichment on for types, descriptions and summaries, or off for a cheaper raw bundle.
Set stale after (days) so every concept carries a refresh date.
Start, then download bundle.zip. Each concept is also a dataset row if you would rather use the API.

Input (paste into the JSON tab)

{
  "startUrls": [
    {
      "url": "https://docs.example.com/api"
    }
  ],
  "maxPages": 200,
  "aiEnrichment": true,
  "staleAfterDays": 90
}

What you get back

{
  "conceptId": "api/create-order",
  "type": "API Endpoint",
  "title": "Create an order",
  "description": "POST /orders creates an order and returns its id and status.",
  "tags": [
    "orders",
    "api",
    "post"
  ],
  "url": "https://docs.example.com/api/create-order",
  "aiEnriched": true
}

Trimmed example. Names and numbers are illustrative; your run returns real data with more fields.

Tips

The bundle follows OKF v0.2: concepts are status: draft and unverified until a person reviews them, which is what the spec expects for generated content.
Links between crawled pages are rewritten to bundle paths like /guides/install.md, so the graph works offline.
Pick an Apify dataset in the input to get a schema table with types, fill rates, AI field notes and joins.

What it costs

$2 per 1,000 concepts without AI, $6 per 1,000 with AI, plus $0.005 per run. There is no page limit.

Plus a small per-run start fee. Apify's free plan includes monthly credit, enough to test. Check the listing for current pricing.

FAQ

What is the Open Knowledge Format (OKF)?

An open spec Google Cloud published in 2026 for packaging knowledge for AI agents: a folder of markdown files with YAML frontmatter where only type is required. Version 0.2 adds provenance, trust and freshness fields.

OKF vs llms.txt?

llms.txt is one index file of links. An OKF bundle holds the knowledge itself, one typed, linked concept per page. This actor outputs both.

Is the output valid OKF?

Yes. Every concept has parseable frontmatter with a non-empty type, index.md files follow the spec's section format, and the root index declares okf_version 0.2.

LLM Wiki & OKF Bundle Generator Turn any docs site or dataset into an LLM wiki your AI agent can browse. $2 / 1,000 pages, $6 with AI.
How to check if ChatGPT mentions your brand
How to turn a website into data for a chatbot
How to use a scraper MCP server in Claude
How to turn a website into an LLM wiki
All How To guides

How to create an OKF bundle from a website