OpenCues never renders explanations inside your text: extra information lives in a secondary display, typically the status line under the input. It shows the highlighted word's tip and its cycle position (word (2/5) - tip), and the active agent task when one is running ([task: correct spelling]).
Why a second surface
The text you're writing is yours; a tip injected into it would move your words around. The dim marks that a word has something to offer; the secondary display says what, only when you navigate to it. Ignore both and nothing interrupts you.
Per-host rendering
Each integration decides what its secondary display is: Claude Code uses its status line, Chrome renders a small status bar, terminals print below the input. The runtime exports the same state everywhere (current word, tip, cycle position, agent task), so behaviour is identical.
Controlling it
tips-mode: off suppresses tip text entirely, and voice-mode: active additionally speaks tips aloud on navigation for eyes-free use.
Related: What is a cue? · What is an inline agent? · What is a kata? · GitHub repo