We type for a living, and we still built this. Every draft, every commit message, every email in this business can start as speech — because the fastest keyboard is the one you don't touch.
Try it live at productivevoice.com ↗ProductiveVoice
Speak. It lands in whatever you're working in — cleaned up, correctly formatted, ready to send.
The Problem
Dictation tools have been "nearly there" for a decade. The gap was never raw accuracy — it's everything around it: transcripts that arrive as one breathless run-on sentence, tools that only work inside their own window, and the copy-paste shuffle that costs more time than typing would have. Voice input only wins when the words arrive where the work is, in the shape the work needs.
- —The wrong window: most dictation lives in its own app; your work lives everywhere else.
- —Raw transcripts aren't writing: "um, so basically" doesn't belong in a client email.
- —Trust: speech is sensitive input; keys and history need proper protection.
What We Built
A voice-first dictation assistant for Windows, designed around one interaction: press, speak, and the finished text appears in your active window.
- —Pill overlay — a small always-available control that floats over whatever you're doing. Trigger it, talk, done.
- —Streaming speech-to-text — live transcription over a Deepgram WebSocket stream, so text forms as you speak rather than after you stop.
- —AI refinement — an LLM pass (via OpenRouter, model of your choice) turns spoken thought into written prose: punctuation, paragraphs, false starts removed, tone preserved.
- —Native paste, any app — Win32 integration places the finished text directly into the focused window: email, documents, code editors, chat. No clipboard gymnastics.
- —A real state machine — dictation flows through an explicit idle → listening → finalising cycle, tested as a pure unit, so the app can't wedge itself between states.
- —Encrypted at rest — API keys and settings are AES-256-GCM encrypted; a history view keeps your dictations searchable, locally.
The Discipline
- —Unit-tested and end-to-end tested — Vitest for the core, Playwright driving the real app.
- —In daily use by us — release notes, emails and documentation at AI Boutique start as speech through this tool.
- —Honest maturity: in active development. The core dictation loop is solid; the edges are still being sanded. It's live at productivevoice.com and we ship updates continuously.
Delivery timeline
state machine, Deepgram streaming, Win32 paste path.
OpenRouter integration, per-context formatting, vocabulary and phrase support.
onboarding, billing, the productivevoice.com launch surface.
What it proves
This is a consumer-grade AI product — streaming, native OS integration, encrypted storage, test coverage — built by the same small team clients hire for automation work. We don't resell wrappers; we build the machinery.
We build speech and automation tooling that fits how your team already works.
Book a Reality Check →