Asking AI with F9

The basics

• Hold F9 and ask in plain speech — no syntax, no wake word
• Release. The overlay appears beside your current window and streams the answer
• Copy what you need, or dismiss and keep working
• Double-tap F9 to open the existing chat panel without asking anything new

The overlay

Answers render with markdown, tables, and math — so a comparison arrives as a table, not a paragraph. Code blocks carry a Copy button, math renders via KaTeX ($$...$$ and $...$), model <think> blocks are stripped with a Thinking… placeholder while streaming, and Ctrl+C copies the answer. The overlay never steals your cursor: your app keeps focus the whole time — click into the panel when you want to select text. Esc closes it.

Overlay layout

Default is docked right at 420×720. Settings → F9 Overlay Position offers Left, Right, Top, and Bottom (720×420) presets, or Free mode: drag the header to move, drag edges to resize (300×260 min, 1200×900 max), with position auto-saved on a 400ms debounce.

Continuous mode

Enable continuous mode to carry conversation context across follow-ups. Ask “explain this error”, then “now show the fix in Python” — the second answer knows about the first. With Chat ON, holding F9 resumes the last session if it is under 30 minutes old (full history re-sent); the history drawer (max 80 sessions) lets you jump threads, and f9-new-chat resets to a blank slate. Turn it off for one-shot questions where prior context would pollute the answer.

Saved chats

Every F9 session is saved locally with full formatting. The Chat History page searches smart — typo-tolerant, weighing questions 3×, follow-ups 2×, answers 1× — with filters and a modal thread view for full conversations. The model behind F9 comes from your per-task assignment, and its system prompt is editable in Settings (default: “You are Flunkey, a helpful voice-activated AI assistant… Keep responses short unless specified and minimize table formatting.”). Wiping everything is one click: Settings → Wipe all data.