A previewer that does not upload your draft
Markdown is the default format for READMEs, documentation, blog drafts and release notes, and the usual way to check how one renders is to paste it into a website. For an unpublished post, an internal runbook or a security advisory being drafted, that is a poor idea. Here the parsing and rendering happen in your browser: the text stays in the tab.
What is rendered
GitHub-flavoured markdown: headings, emphasis, strikethrough, ordered and unordered lists, task lists, tables, blockquotes, links, images, inline code and fenced code blocks. Line breaks are treated as breaks, matching how most editors and issue trackers behave, which is usually what surprises people when their carefully wrapped paragraph renders as one block elsewhere.
Footnotes, definition lists and the various directive syntaxes are extensions rather than core GFM and are not rendered.
Copying HTML versus saving markdown
These are two different jobs, so there are two buttons. Copy puts the rendered HTML on your clipboard — the form you need when pasting into a CMS, an email client or a rich-text field that does not understand markdown. Download saves the markdown source as a .md file, which is what you want for the repository.
Writing markdown that survives the trip
Renderers differ. If a document has to look right in more than one place — GitHub, your docs site, and a CMS, say — stay close to the common subset: headings, lists, links, code fences and tables. Raw HTML embedded in markdown is the most frequent cause of a document that renders beautifully in one viewer and falls apart in another, since many renderers strip it for safety.