Features

Formatted text at your cursor.
Audio that never leaves your accounts.

Productive Voice is built on two core principles: it should never break your flow, and it should never see your data. Every feature below is an answer to one of those two questions.

Ctrl+Shift+Space anywhere Floating pill 5 style snippets Custom dictionary Your Deepgram API key Your OpenRouter API key Local-only history
⟢ USP 01: Workflow

Speak once and get formatted text typed at your cursor, without ever leaving the app.

Most dictation tools steal focus. They open a window. They make you copy and paste. Productive Voice doesn't. It floats over whatever you're doing, takes your voice, and types the result exactly where your cursor was.

One key. Anywhere.

The hotkey is the whole UI.

No menu bar dance. No window to find. Whatever app you're in (VS Code, Outlook, Notion, Figma), press the combo and Productive Voice takes over the airspace above it.

Default Ctrl Shift Space · fully rebindable
Push-to-talk or toggle modes
Per-app overrides (Slack = casual, Outlook = formal)
Ctrl
+
Shift
+
Space
Storyboard · 0.0s → 8.5s
Status: Idle
01 Trigger 0.0s
vs code · main.py
def handle_request():
  # TODO:
Ctrl Shift Space

From any app. Your active window keeps focus: Productive Voice's pill appears above it.

02 Speak 0.1 – 8s
4.2s

Real-time transcription via Deepgram, streamed direct under your API key.

03 Format + 0.3s
Deepgram → transcript 142ms
OpenRouter → format ~280ms
Style: email_formal queued

An LLM under your OpenRouter key shapes raw words into the right output for the active app.

04 Paste + 0.1s
outlook · new message
To: sarah@
Hi Sarah,
Could you send me the Q2 numbers when you get a chance?

Productive Voice types the formatted text exactly where your cursor was. The pill fades. You keep going.

Total elapsed
≈ 8.5 s
Architecture

Two-stage pipeline.
Each model doing what it's best at.

Deepgram knows speech. LLMs know language. We don't try to make one model do both, and that's why Productive Voice feels fast.

Input
Your mic
PCM 16kHz
Microphone Capture Saves audio locally in high-fidelity 16kHz PCM mono format. Fully buffered for zero-latency transmission to APIs.
audio
Stage 1 · Transcription
Deepgram Nova-2
~140ms · streaming · your key
Nova-2 Transcription Ultra-low latency streaming model. Direct websocket connection under your personal Deepgram API key. Direct & secure.
text
Stage 2 · Format
OpenRouter LLM
~280ms · gpt-4o-mini · your key
Formatting Phase Contextual formatting using models like GPT-4o-mini or Claude-3-Haiku. Strips filler words, handles punctuation, and structures grammar according to active style snippets.
paste
Output
Your cursor
Any app, any field
Native Injection Uses safe OS keyboard event simulation to directly paste or type formatted text. Compiles perfectly in VS Code, Slack, Outlook, Notion, etc.
Feature 002 · Style snippets

The same dictation, in five different voices.

Pre-built styles for Email, Slack, Bullets, Code comments, and Notion ship by default. Each is a prompt you can rewrite in seconds. The output below is real: same raw transcript, four different style snippets applied.

  • 5 styles ship by default, fully customisable
  • Per-app default style (Slack defaults to casual, Outlook to formal)
  • Switch styles with Ctrl 1–5 mid-dictation
  • Your prompts run under your own OpenRouter key
Email
Slack
Bullets
Code
"hey sarah uh could you send me the q2 numbers when you get a chance also let's sync about the launch on thursday i was thinking maybe two pm"
email_formal · gpt-4o-mini

Subject: Q2 numbers + Thursday launch sync

Hi Sarah,

Could you send me the Q2 numbers when you get a chance? Also, let's sync about the launch on Thursday, I was thinking 2pm.

Thanks!

Without dictionary With dictionary →
RAW
"Let's deploy this to cuber-nettys on post-grease-cool"
FIXED
"Let's deploy this to Kubernetes on PostgreSQL"

Custom dictionary

28 terms
Kubernetes koo-ber-NET-eez ×
Anthropic an-THROP-ik ×
PostgreSQL POST-gres-ess-cue-el ×
Tanaka-san tah-NAH-kah-san ×
Phonetic Correction Sandbox

Type raw sound-alikes (e.g. cuber nettys, post grease cool, or your added terms) below to see them correct in real-time.

Corrected live transcript appears here...
Feature 003 · Custom dictionary

Productive Voice learns the words that matter to you.

Names, acronyms, technical jargon, proper nouns: anything Deepgram doesn't get right by default. Add it once. Productive Voice boosts it through the Deepgram API and the LLM cleans up the rest.

  • Optional pronunciation hints for tricky terms
  • Synced across Deepgram boost + LLM prompt
  • Lives encrypted on your machine. Never synced to us.
⟢ USP 02: Privacy

We can't see your voice.
Even if we wanted to.

Bring-Your-Own-Key isn't a marketing checkbox. It's the architecture. Your audio goes from your microphone straight to Deepgram, under your account. We're not on that wire. We don't have a server you talk to.

PILLAR 01

Bring your own keys

You hold the Deepgram and OpenRouter accounts. All audio and transcripts flow under your credentials, billed to your account, governed by your terms with those providers.

PILLAR 02

AES-256-GCM at rest

API keys, dictation history, and dictionary entries are encrypted on disk using OS-keychain-derived keys. Even raw disk access won't expose them.

PILLAR 03

Zero server-side storage

No account. No telemetry. No transcripts. No logs. We literally cannot answer the question "what did this user dictate?", even under court order.

PILLAR 04

Local-only history

Your dictation history is stored encrypted in your user folder. Searchable, replayable, exportable, but only by you, on your machine, on this OS install.

vs. Others

How Productive Voice compares to the alternatives.

Built-in Windows dictation is free but limited. Subscription tools charge forever and store your audio. Productive Voice is the third option.

Feature Comparison
Productive Voice
Wispr Flow
Aqua Voice / Otter
Price
$39 one-time
$12/month (£9.50)
$10–30/month subscription
AI Formatting & LLM Choice
Yes, choice of LLM
Yes, fixed provider
Fixed / basic formatting only
Bring-Your-Own API Keys
Yes, Deepgram + OpenRouter
No
No
Audio Leaves Your Accounts
No, direct to your accounts
Yes, sent to vendor
Yes, sent to vendor
Works in any Windows app
Yes, floating pill overlay
Yes
Mostly (some layout quirks)
Custom Dictionary
Yes, unlimited + local storage
Yes
Paid / team tier only

Comparison reflects publicly available pricing and features. All active subscription products require ongoing monthly fees.

Ready when you are

Hand the typing to your voice.

$39 one time. 14-day refund. Lifetime updates.