TL;DR

a sends the current note (body without front matter and %%comments%%, cut to maxChars); the 2–3 sentence summary streams in at the top of the Info tab as a footnote: *, the text, and a rule signed TL;DR · <model>.

• Up to 14 rows show; a longer one ends ⏎ more, and ⏎ on its rule shows the rest (⏎ again folds it).
• While the model works the block says asking <model> · 3 s; an older summary stays dimmed under it until the first words arrive.
• Summaries are cached and shown again for free.
• One made before the note changed is dimmed with edited since · a refresh.
• A failed request leaves its error in the block with a retry (and in the toast) until you retry, open another note, edit the note or press r.
• A * after the title in the top line says the note has a TL;DR (muted when it is outdated); the document footer offers a summarize until it is fresh.
• One that lands while Info is out of sight says TL;DR ready · 3 to read, or names its note when you have moved on.
• One request per note; a on a running note cancels it.
• Without an ai block, a says AI off · set "ai" in config.json.
• For a piped document the summary lasts for the run and is not cached.

The ctrl+p peek pane shows a fresh cached TL;DR without sending anything.