Speak before choosing
Intent is inferred after transcription, so the person does not have to predict the correct mode before they know what they want.
Native macOS interactionSay it. Shape it.
Say it. Shape it.
Ship it.
Voicy turns speech into polished writing without reading the screen, stealing focus, or acting before the user approves.
⌥Space hold to speakWhat should Voicy make?Transcript ready
“Can you send Sarah an update about the proposal and ask if Thursday morning works for kickoff?”
Subject: Proposal update and kickoff
Hi Sarah, I wanted to share an update on the proposal and check whether Thursday morning works for our kickoff.
Hi Sarah, I wanted to share an update on the proposal and check whether Thursday morning works for our kickoff.
ListeningRelease Option + Space when finished
Product thesis
Voice input should feel like a layer, not another destination.
Most dictation tools add modes, history, settings, and context capture around a small writing job. Voicy stays invisible until needed, then expands only as the decision becomes richer.
Preview before effect
Every transformation becomes editable text. Nothing is inserted, sent, or executed without explicit approval.
No hidden context
Version one does not read the screen, clipboard, selected text, files, or conversation history. The transcript is the input.
Core experience
From invisible to useful in four states.
Motion preserves spatial continuity. The surface expands upward from a bottom-center capsule and never becomes a full application window.
STATE 01
Rest without visual noise
A 52 by 8 point bottom-center hotspot, the menu bar, and the keyboard shortcut keep Voicy reachable. At rest, the desktop shows no visible pixels.
STATE 02
Hold to capture
Option + Space opens a focused listening capsule with waveform, elapsed state, Stop, and Escape. The passive surface does not steal keyboard focus.
●Listening0:07 · Release to stop
STATE 03
Understand the writing job
The transcript appears before analysis finishes. Up to four allowlisted actions are suggested, and one may be recommended but never auto-run.
Clean transcriptEmailProfessionalShorterTranslateFix grammar
STATE 04
Edit, approve, insert once
Voicy revalidates the original target before one physical insertion attempt. Uncertain outcomes remain visible and never retry automatically.
READY FOR REVIEW
Hi Sarah, I wanted to share an update on the proposal and check whether Thursday morning works for our kickoff.
⌘Return approve insertionTrust boundary
Clear about what moves, where, and why.
The experience names each provider state and limits the system to text transformations. Useful automation begins with disciplined constraints.
On-device firstAudio capture and target resolution begin locally. Capture ends completely before any provider upload.
Keychain-held credentialThe owner’s API key is non-synchronizing secret material and never enters logs, evidence, or accessibility text.
Text actions onlyEmail means a subject and body, not send. AI Prompt means a structured request, not execution.
One safe commitTarget revalidation happens immediately before insertion. Ambiguity stops the flow.
“Turn this into a concise, professional project update.”Technical tokens, names, and meaning stay intact.
«حوّل هذا الكلام إلى تحديث مشروع واضح ومختصر.»اتجاه صحيح، علامات ترقيم مناسبة، وحماية للمصطلحات التقنية.
System designSmall surface.
Small surface.
Serious state machine.
Native AppKit presentation sits above an explicit session coordinator, bounded provider contracts, and fail-closed insertion behavior.
01Global shortcut and content-free target token
02Bounded in-memory audio capture
03Transcription and closed-schema intent analysis
04Allowlisted text transformation
05Editable preview and target revalidation
06Single confirmed insertion attempt
11versioned text actions
10 minbounded capture duration
24 MiBmaximum encoded audio payload
EN + ARfirst-class language behavior
Product reflection
Product strategy · Interaction design · macOS architectureLocal private prototype · Not a public release
Magic is better when its boundaries are visible.
Voicy’s differentiator is not speech recognition alone. It is the choreography of trust: reveal at the right moment, explain the state, preserve control, then disappear.
Product strategy · Interaction design · macOS architectureLocal private prototype · Not a public release