If you want to feel out a structured-extraction API before you point it at something hard, GitHub is a good place to start. Repository pages render their important metadata server-side, the content is stable, and the layout is consistent across millions of pages. That combination is exactly what makes an extraction predictable.
A repo schema
Here is a schema that captures the fields most people actually want off a repository page:
const repo = await client.extract({
url: "https://github.com/facebook/react",
schema: {
name: z.string(),
owner: z.string(),
description: z.string(),
primaryLanguage: z.string(),
stars: z.number(),
topics: z.array(z.string()),
},
});You send that, and you get back a clean object. No parsing the star count out of a string like “228k”, no guessing which heading is the description. The numbers come back as numbers because the schema said so.
Why this one is easy
A page is easy to extract from when three things are true, and a GitHub repo page hits all three:
- It renders server-side. The metadata is in the initial HTML, so it is available the moment the page loads.
- It is consistent. Every repo page has the same shape, so one schema works across all of them.
- It is cooperative. GitHub is not throwing anti-bot walls at a normal request for a public page.
When you are learning what a tool can do, start on pages like this. You get a clean signal about extraction quality without fighting infrastructure.
Scaling it up
Once the single-repo extraction works, the same schema runs across a list of repos. Feed it the URLs from an org, a topic page, or a starred list, and you get a uniform dataset where every row has the same fields and the same types. That is the part that makes downstream work pleasant: you are not writing defensive code for the one repo that formatted its star count differently.
A note on what is hard
Not every site behaves like GitHub. Pages that render everything client-side, sit behind aggressive anti-bot systems, or change layout constantly are a different conversation. We are honest about that distinction, and we think you should pick your early targets accordingly. Cooperative, server-rendered, structured pages first.
Try it
The live demo on the home page defaults to a GitHub repo for exactly this reason. Run it and change the URL to any repo you like, or start free to script it.