The original note, as written at the time
As you input text you can attach text input to each bite of text. Sometimes you can draw or sketch something in the input panel and attach a sound bite as well.
Ongoing development
As you input text you can attach an input to each part of it — a sketch drawn in the panel, a sound bite, a note in the margin.
Writing is not only words, and we have spent thirty years building editors that pretend it is. When a person is drafting something difficult, the useful material is often not a sentence: it is a diagram, a voice memo made in the car, a photograph of a page in a book.
Today those things live in different applications and the connection between them is a filename nobody will ever find again. The text and its evidence are separated at the moment of creation, which is the moment the connection is worth most.
Attaching to a specific passage rather than to the document is the important part. A sound bite hanging off a paragraph is a note to the author about that paragraph. A sound bite attached to the file is a file in a folder, and folders are where things go to be forgotten.
The interface should be invisible when it is not wanted, which is the hard part. Every editor that added annotations made the page busier, and a page that is busy is not a page you can think on. So the attachments should be indicated by the smallest possible mark in the margin, and the material itself should open over the text rather than beside it.
The real value appears years later, when a draft is reopened and the voice memo explains what the paragraph meant before it was polished into something that no longer says it.