Hyperapp
The capability unit of Hypernet. A website whose first reader is an agent.
A Hyperapp is the capability unit of Hypernet: a website whose first reader is an agent. A website is to the internet what a Hyperapp is to Hypernet.
The page is the interface
A Hyperapp exposes everything through the page itself. Reading the page is knowing the current state. Links are discovery: every capability is reachable by following them from the root. Forms are actions: submitting one does the work. There is no schema beside the page and no documentation behind it. What the agent reads is what the app can do.
One source, two representations
A page is authored once and served twice. A human receives HTML and sees an
ordinary, working page. An agent receives Markdown of the same page by
content negotiation (the Accept header), rendered from live state at
request time (see
Dynamic Context Rendering). The
capabilities are identical in both representations, and because they come from
one source, they cannot drift.
Built agent-first
A Hyperapp's information architecture serves its reader. An agent reads sequentially, pays for every token, and walks trees instead of scanning dashboards. So a Hyperapp keeps its pages short, gives each page one concern, names its actions stably, and puts every capability a link away from the root. It does not carry a human product's layout, and it does not need to.
Fronting a service
Many Hyperapps front a third-party service: the app is the service's client, and the agent acts on the service through it. Two host-mediated pieces complete the picture. A connection is the agent's authorized access to the service: a human grants it once, and it stays bound to that agent. A trigger turns the service's events into moments that wake the agent, so work flows in both directions. Both are optional. A pure capability (a calculator, a slide builder) needs neither.
Where humans meet it
Humans do not browse Hypernet. Their agents do. The conversation is the human's window, and the Hyperblock is the opening in it: an Embedded View projects the app's rendered output into the conversation, and a Hyperform collects a human's input for a step that needs it. The pages themselves stay legible to people. The author debugs in a browser, machines parse the structure, and an audit can open exactly what an agent saw. Human action flows through the blocks. So does human-grade fidelity.
The full model
The specification is normative: the page and Action rules, the Response Page contract, and the Markdown representation. Hyperkit is the toolkit that builds to it.