agentkit-web

Expose agent-readable, authoritative website data for AI agents — without scraping, SEO hacks, or UI parsing.

agentkit-web lets you publish a canonical truth endpoint that AI agents can rely on deterministically.

What this solves

$3

* Scrape HTML
* Guess intent
* Hallucinate availability, pricing, or policies

$3

* Don’t expose machine-readable truth
* Are optimized only for humans and search engines

agentkit-web bridges this gap.

Core concept

Every website exposes a single, well-known endpoint:
/.well-known/agent.json

This endpoint declares:
* Who you are
* What data is authoritative
* What constraints exist
* Where structured truth lives

Agents fetch this first. Search becomes a fallback.

Installation

``bash
npm install agentkit-web
`

Usage

$3

`typescript
import { defineSite, serveAgent } from "agentkit-web";

const site = defineSite({
site: "example.com",
description: "Example agent-readable website",
endpoints: {
products: "/agent/products.json",
policies: "/agent/policies.json"
},
constraints: {
delivery_available: false
}
});

export const GET = serveAgent(site);
`

This exposes:
GET /.well-known/agent.json

$3

`json
{
"schema_version": "0.1",
"site": "example.com",
"authority": "official",
"description": "Example agent-readable website",
"last_updated": "2026-01-12T12:00:00Z",
"endpoints": {
"products": "/agent/products.json",
"policies": "/agent/policies.json"
},
"constraints": {
"delivery_available": false
}
}
`

What this is NOT

* Not SEO
* Not schema.org
* Not a crawler
* Not a search engine
* Not a ranking signal

It has zero impact on UI or search rankings.

Design principles

* Deterministic over probabilistic
* Intent-based, not page-based
* Minimal and flat JSON
* Zero UI coupling
* Zero SEO coupling

Works best with

* AI agents
* RAG pipelines
* Autonomous assistants
* Internal enterprise AI systems

For agent-side consumption, use agentkit-agent`.

Versioning

* 0.x — Schema evolving
* 1.0.0 — Schema locked

Current version: 0.1

License

MIT