Structured web context for AI agents

Agents got smarter, but they still can't really see the live web. Drop a raw HTML page into a context window and you've spent thousands of tokens on navigation chrome, cookie banners, and markup before the model reaches a single useful fact. The fix isn't a bigger context window — it's better context.

The shape of good context

Good agent context has three properties. It's fresh (minutes old, not months), structured (typed fields, not prose to re-parse), and scoped (only what the task needs). The Hog is built around delivering exactly that.

Here's how a few common signals map to where they come from:

A worked example

Say your agent qualifies inbound leads. Instead of handing it a homepage, hand it a structured record:

const company = await hog.enrich.company({ domain: "example.com" });
 
if (company.headcount > 50 && company.hiring) {
  await routeToSales(company);
}

The agent reasons over clean fields — headcount, hiring, funding — and never touches a line of HTML.

Keep the loop tight

A few principles we've found hold up in production:

1. Fetch narrowly. Ask for the fields the task needs, nothing more.
2. Cache aggressively. Most context is reusable across runs within a session.
3. Monitor, don't poll. Subscribe to changes instead of re-scraping on a timer.

The best agent context is the smallest set of fresh, structured facts that lets the model make the next decision.

That's the bar we hold every endpoint to.