A custom HTML page is a resource with type: hm://hyper.media/page. Its html attribute points to an IPFS folder containing index.html; its inputSchema points to a published Hypermedia schema. This is a page, rather than a larger application.

The input schema is an HM struct: one property for each input field. Publish the schema blob and optionally bind it to a readable type resource through schemaDefinition. Other tools can discover the contract by reading the page's inputSchema reference. A page without inputs uses page-input.

Input lives in the page URL. Each field name becomes the query parameter p.name, whose value is JSON encoded and then URL encoded. For example ?p.prompt=%22Hello%22&p.count=3. Seed's own version and panel parameters are separate. Input is public URL state: do not place secrets in it.

The SDK provides seed.input.get(), seed.input.push(values), seed.input.replace(values) and seed.input.onChange(listener). Push adds a history entry; replace updates the current one. Both take the complete input object and validate it against the declared schema before changing the URL. Browser and desktop Back/Forward restore the input without restarting the page. Invalid incoming URLs are reported as input errors so the page can show a correction UI.

See custom HTML pages for sandboxing, permissions, publishing and schema-backed data storage.