Unique Feature

Dictation

Talk to your DA with a foot pedal: tap for long thoughts, hold for quick ones.

A machined foot treadle hinged on a steel pin at one end of a long graphite base, pressing a smoked-glass cylinder; a violet channel runs past three stacked black strata and a lifted pawl into a ring terminal, the far end cut away into bone-grey outline.
Fig. 19Press to speak, release to send

Dictation is how you talk to your DA instead of typing to it. A foot pedal and a keyboard shortcut drive a dictation app, Typeless, in two ways. Tap once to start talking and tap again to stop, for a long hands-free ramble. Or hold the pedal while you speak and let go, and your words are pasted into the conversation and sent.

Why it exists

The quality of what your DA does depends on how much of your intent reaches it. Typing is a narrow pipe for that. You shorten the request, drop the context you would have mentioned out loud, and skip the “and also” that was the most important part. Speaking carries all of it, and it is faster than typing.

Most dictation tools get in the way at the moments that matter. One mode means you either reach for the keyboard to stop every short burst, or you hold a key for a two-minute explanation. Neither fits how you actually talk to an assistant, which is a mix of quick corrections and long thinking-out-loud. Dictation gives each its own gesture on the same pedal, so your hands stay on whatever they were doing.

How it works

The pedal is a Stream Deck foot pedal running a small custom plugin. The plugin times each press. A press shorter than 0.4 seconds is a tap: it toggles dictation on or off, and nothing else happens. A longer press is a hold: dictation starts when your foot goes down and stops when it comes up, and then the plugin waits for the text to land and presses Enter for you.

Three parts of that took more than a settings change, and they are the reason the feature is documented as a system instead of a preference:

  • the hold→ Typeless only supports tap-to-toggle and refuses a held shortcut. So the plugin fakes a hold: one tap when you press, a second tap when you release. emulated
  • the key→ Stream Deck's built-in hotkey action sets a Control flag without pressing a Control key, and Typeless tells left and right Control apart, so it ignores that. The plugin posts a real right Control key around the shortcut. plugin
  • the binding→ Typeless is bound to Right Control plus F19, a combination no keyboard sends. That leaves the easy shortcut, Right Control plus J, free for a small Hammerspoon handler that gives the keyboard the same tap-or-hold behaviour as the pedal. keyboard

The auto-send is careful about when it fires. After a hold ends, the plugin reads Typeless’s own local history to see how that dictation finished. It presses Enter only when the text was actually pasted. A cancelled dictation, one with no speech in it, or one that takes longer than twenty seconds gets no Enter, so a false start never sends half a sentence.

Whatever is playing gets out of your way while you talk. When a dictation starts and a browser, Music or Spotify is playing, the plugin pauses it, and it plays again when the dictation ends. It pauses rather than mutes on purpose. Many pro-audio interfaces have no mute that macOS can flip, and muting around them means reconfiguring the Mac’s audio on every press, which made the whole machine stutter and made Typeless miss the release. Pausing only presses the media key, so nothing else is disturbed. It follows Typeless’s own “mute audio when dictating” switch.

One held burstfrom your foot to a sent message
PressYour foot goes down and Typeless starts listening.
→
ReleaseAfter more than 0.4 seconds, letting go stops it and Typeless pastes the text.
→
SentThe plugin sees the paste complete in Typeless's history and presses Enter.
A tap never sends. Toggle mode is for long thinking, so it pastes and waits for you to read it over before you send it yourself.

Because none of this would survive a new machine by memory, it lives in LifeOS as one feature. The documentation explains every piece and why it exists. One setup command rebuilds it on a fresh Mac, and a check command names the broken piece in one line when something stops working.

Where it fits

Dictation is the input half of talking to your DA, and Voice is the output half. You speak a request with the pedal, the DA works, and it speaks back what it did. Between them the terminal becomes a conversation you can have while looking somewhere else.

It also changes what reaches the rest of the system. A spoken request carries the reasons and the constraints you would have cut from a typed one, and that is the raw material the DA turns into ideal-state criteria. More of your intent goes in, so less of it has to be guessed.

What it feels like

You lean back and think out loud for two minutes with one tap at each end, and the whole ramble lands in the conversation for you to read over. A minute later you hold the pedal, say “yes, and run the tests too,” and let go, and it is already sent. Your hands never went to the keyboard, and the requests are longer and clearer than the ones you used to type.