Voice
Voice is an entrance, not a place. You speak, Milo listens, understands, works and is done — then the pill disappears. It is the same Milo Core the composer talks to, aware of the surface you are on.
The pill
A transient pill at the top centre: ring, state word, live transcript. Nothing else.
- Ring — Milo Vermilion. Its motion is the state: breathing, dashed rotation, pulse, resolve.
- State word — mono microtype. The label carries the state; the ring only echoes it.
- Transcript — live, interface sans, one line, ellipsised. Replaced by Milo's reply in Working and Done.
- Position — fixed, top-centre, 16px from the edge, level 3 shadow. It never covers the primary action.
Five states, then gone
Listening, understanding, working, done, error — each a change of ring and word, then the pill leaves.
Try it
The full sequence runs here; Quick Entry is the keyboard equivalent.
Voice
Runs the full sequence at the top of the viewport, then disappears.
Quick Entry — the keyboard equivalent
Invoke. Type. Submit. Disappear. The toast is Milo's acknowledgement.
Invocation
Hold a key, speak, release; the current surface is the context.
Invocation rules
- 01One entrance from anywhere: the system hotkey, the menu-bar item on macOS, the Hyprland bind on Omarchy, or the ◉ button on a surface.
- 02Milo already knows the context — the current project, task, artifact and machine. You do not name them unless you mean a different one.
- 03Anything that changes files, spends money or touches another machine goes through a PermissionRequest first. Voice never bypasses level 4.
- 04Milo replies in UI language: brief, calm, precise. "On it." "Done." "This needs your approval."
- 05Under reduced motion the ring is still; the state word does all the work.
What voice is not
Not an orb, not a waveform, not a dashboard, not a second assistant.
- A Siri orb
- A giant waveform
- A voice dashboard or history surface
- A separate assistant with a separate memory
- Always listening