Hot Mic

Hot Mic is Field Theory's always-on voice layer for command-first, hands-free workflows. If push-to-talk is for intentional capture, Hot Mic is for continuous flow: speak naturally, then trigger submit, paste, or action phrases.

Hot Mic is very experimental and actively improving. Please @andrewfarah on X if you have feedback.

---

Requirements

• macOS 14+
• Apple Silicon (M1 or later)
• Qwen transcription engine selected in Field Theory
• Qwen voice model installed (open source, runs fully offline)
• Microphone and accessibility permissions granted

Quick Start

1. Open Settings -> Audio & Transcription and select Qwen.
2. Install Qwen if prompted.
3. Open Settings -> Hot Mic and enable Hot Mic.
4. Speak naturally.
5. End with a submit phrase (for example, go ahead) to send.

---

Default Voice Commands (Editable)

Each group below is editable in Settings -> Hot Mic.

1) Dictation Control

• Submit buffered text: go ahead, send it, submit, do it
• Paste buffered text without submit: paste, paste it, transcribe
• Cancel/interrupt (clears buffer): stop, abort

2) Window and Terminal Workflow

• Next window: next window, switch
• Previous window: previous window
• New window: new window
• Close window: close window, close the window, close this window
• Minimize: minimize, minimize window, minimize the window
• Hide app: hide, hide app, hide this app, hide the app
• Quit current app shortcut: quit app, quit this app
• Start Claude: start claude, start cloud, run claude, start clod
• Start Codex: start codex, run codex
• Restart server: restart server, restart dev, restart dev server

restart server requires a configured command (for example, npm run dev).

3) Media and System Controls

• Play/pause: play, pause, play pause, play music, pause music
• Next track: next track, next song, skip song
• Previous track: previous track, previous song, go back a song, last song
• Volume up: louder, volume up, turn it up
• Volume down: softer, quieter, volume down, turn it down
• Mute: mute, mute audio
• Unmute: unmute, unmute audio
• Sleep Mac: go to sleep, sleep computer
• Lock screen: lock screen, lock computer

4) App Switching and App Quit by Name

Hot Mic supports phrase prefixes plus app names and aliases.

• Open/switch prefixes: open, switch to, go to
• Hide-by-name prefix: hide
• Quit prefixes: quit, close, kill

Examples: open chrome, switch to terminal, go to cursor, hide slack, quit slack, close spotify.

5) Window Layout Commands (Squares)

• Grid/tile: grid, tile, tile all, grid all
• Show windows: show all, show all windows, show windows
• Focus mode: focus, focus mode, center focus, hide others, hide other windows
• Layouts: horizontal, spread horizontal, side by side, vertical, spread vertical, stack windows, cascade, cascade windows
• Position: snap left, snap right
• Corners: top left corner, top right corner, bottom left corner, bottom right corner
• Screen/restore: maximize, full screen, fullscreen, enter full screen, exit full screen, leave full screen, center, center window, restore

6) Portable Command Files

• Single command file: use the <name> command
• Multiple command files: use the commands <a>, <b>, and <c>

7) Fast Reply Shortcuts

• first option -> 1
• second option -> 2
• third option -> 3
• fourth option -> 4
• allow or approve -> y
• always -> a
• deny -> n

---

Context Stacking

• Hot Mic currently focuses on voice dictation and voice commands.
• Screenshot context stacking is not supported yet in Hot Mic mode.
• For now, take screenshots and paste them manually.
• In terminals like Claude Code or Codex, use Command+Shift+V (Super Paste) for images. This pastes image file paths the terminal workflow can reference.

Customize and Reset

Everything is editable in Settings -> Hot Mic, including submit/paste/cancel phrases, app prefixes, system command phrases, restart-server behavior, and app voice aliases.

If things drift, click Reset Voice Defaults to restore the default phrase sets.

Customize Voice Commands Outside Default Settings

Modifying default phrase groups is encouraged, but do it carefully.

• Single-word commands are more likely to fire accidentally during normal dictation.
• There is always a tradeoff between fast command triggers and clean conversational dictation.
• You may trigger a command when you meant to keep talking. Hot Mic has guardrails, but mileage may vary.
• Strong recommendation: add phrase-based commands first, then test single-word commands incrementally.
• Avoid changing many commands at once. Small edits preserve flow and make regressions easier to isolate.

Dynamic Island

• The orange dot means Hot Mic is actively listening.
• The transcript dropdown appears after the first five words to reduce early hallucinated fragments.
• Your first five words are still captured. The display is delayed; transcription is not.
• Speak naturally. You do not need to pause or repeat the start of a sentence.
• You can recover recent entries from Dynamic Island transcript history (up to 25).

Hot Mic vs Push-To-Talk

• Hot Mic: ambient, command-driven, continuous buffering.
• Push-to-talk: intentional, bounded by hotkey press/release.

Privacy

Hot Mic command detection and transcription behavior are local-first. Auto-improve is not available in Hot Mic mode.

Safety and Recovery

• Commands are matched at the end of utterances.
• If buffered text exists when a command is detected, Field Theory flushes text first.
• Cancel phrases intentionally discard buffered text.

Voice Tuning (Background Voice Filter)

In Settings -> Hot Mic, use the Background Voice Filter toggle and strictness slider (0-100) when nearby voices cause false triggers. Filtering is off by default and strictness defaults to 4.

• This is an especially hard problem and still an active area of work.
• Please calibrate expectations in noisy environments. Hot Mic may not work reliably when other people are talking nearby.
• If it works well in a noisy space, great. But today, quiet environments are still the most reliable setup.