By Denys MedvedievHow It Works
How Whisper Actually Works
Press a hotkey, speak naturally, get polished text — in any app. Here’s exactly what happens under the hood, with real UI examples.
Last updated: April 2026 · Applies to v2.1+
The Overlay
Whisper lives as a small floating overlay on your screen. It stays on top of all windows but never gets in your way. Here’s how the overlay looks in each state:
Idle — Collapsed
When you’re not interacting with Whisper, the overlay collapses to a tiny pill — barely visible, completely out of your way. It takes up almost no screen space.
Collapsed pill. The overlay stays minimal until you need it.
Idle — Hover to Expand
Move your mouse over the pill and it expands to show your recording hotkey. This is your visual reminder of how to start recording.
Expanded on hover. Shows your hotkey — press it to start recording.
Recording — Listening to Your Voice
Press your hotkey (default: Ctrl + Space) and start speaking. The overlay shows an animated waveform and a red stop button. Release the hotkey or click stop to finish.
Recording in progress. Waveform shows audio is being captured.
Transcribing — Speech to Text
After you stop recording, the overlay switches to a transcribing state. Your audio is being processed — either locally on your device or via OpenAI’s API, depending on your mode.
Audio is being transcribed to text. Takes less than a second for short phrases.
AI Enhancing — Polishing Your Text
If AI enhancement is enabled, the overlay shows a purple “Thinking” indicator. The AI cleans up filler words, fixes grammar, and restructures your sentences into polished text.
AI is enhancing your transcription. Powered by Ollama (local) or OpenAI GPT.
Web Search — Finding Answers
When you trigger a web search (by saying your AI keyword followed by a question), the overlay turns blue as it searches the internet and synthesizes an answer with sources.
Searching the web for real-time data. Answer with sources delivered in seconds.
Complete — Text Delivered
A green checkmark confirms the text has been pasted into your active application. The entire process — from speech to final text — typically takes 2–5 seconds.
Done! Text has been pasted into your active application.
Use Case: AI Text Enhancement
Let’s walk through a real example. You’re writing meeting notes in a text editor and want to quickly summarize what was discussed. Instead of typing it out, you speak naturally and let AI clean it up.
Scenario: Summarizing a Meeting
You just finished a project standup and need to send a quick summary to the team. Instead of typing, you speak freely and let the AI polish it.
What You Say
“so basically um we had the standup and uh john said the API migration is done and it’s in staging now and sarah mentioned that the mobile app has a bug with push notifications on android and she’s looking into it and um we decided to push the release to next wednesday because we want more testing time”
Whisper Records & Transcribes
AI Enhances Your Text
Polished Text Appears in Your Editor
Standup Summary
• API Migration: Complete and deployed to staging (John)
• Mobile Bug: Push notifications broken on Android — under investigation (Sarah)
• Release: Pushed to next Wednesday for additional testing
What Just Happened
You spoke for ~15 seconds with filler words and run-on sentences. The AI removed the fluff, extracted key points, structured them into a clean summary with names and action items, and pasted it directly into your editor. No typing. No copy-paste. No editing.
Before & After
“so basically um we had the standup and uh john said the API migration is done and it’s in staging now and sarah mentioned that the mobile app has a bug with push notifications on android…”
Standup Summary
• API Migration: Complete, in staging (John)
• Mobile Bug: Android push notifications (Sarah)
• Release: Moved to Wednesday for testing
This works with any kind of content — emails, documentation, Slack messages, reports. Speak rough thoughts, get clean professional text.
Use Case: Web Search with AI
Whisper’s web search lets you ask questions by voice and get researched answers with sources — delivered right where your cursor is. No browser, no tab switching, no copy-pasting.
Scenario: Getting Market Data
You’re writing a financial report in a document and need the latest S&P 500 data. Instead of opening a browser, you ask Whisper by voice.
What You Say
“Hey Whisper, what is the current S&P 500 price and how has it performed this year?”
“Hey Whisper” is the AI keyword that triggers the AI assistant instead of plain transcription. You can customize this keyword in settings.
Whisper Records Your Question
AI Searches the Web
The AI searches multiple sources, cross-references data, and compiles a researched answer — typically in about 5 seconds.
Answer Appears in Your Document
S&P 500 — Current Data
The S&P 500 is currently trading at approximately 5,611.85 as of early April 2026.
Year-to-date Performance
The index is up +8.2% year-to-date, building on the strong momentum from 2025. Key drivers include continued AI sector growth, resilient consumer spending, and stabilizing interest rates. The Federal Reserve held rates steady in its March meeting, which boosted market confidence.
Sector Highlights
• Technology: +14.3% YTD (led by AI and semiconductor stocks)
• Healthcare: +6.1% YTD
• Energy: -2.8% YTD (due to lower oil prices)
Sources: Yahoo Finance, Reuters, MarketWatch
What Just Happened
You asked a question by voice. Whisper recognized your AI keyword (“Hey Whisper”), searched the web for current S&P 500 data, synthesized information from multiple sources, formatted it with sections and cited sources, and pasted it directly into your Word document. Total time: ~5 seconds. Zero browser tabs opened.
More Examples
Web search works for anything you’d normally Google. Here are a few more examples:
“Hey Whisper, what are the new features in React 19?”
Get a structured breakdown of React 19 features with code examples — pasted right into your IDE or notes.
“Hey Whisper, compare PostgreSQL vs MySQL for a new project”
Get a researched comparison table with pros, cons, and recommendations — ready to share with your team.
“Hey Whisper, what’s the weather in Tokyo this week?”
Get a quick weather summary for travel planning — pasted into your email or notes app.
The Complete Flow
Here’s the entire process from start to finish:
You’re working in any app
VS Code, Slack, Google Docs, email — anywhere with a text field
Press your hotkey & speak
Hold Ctrl+Space (or your custom key) and talk naturally
Whisper transcribes & enhances
Speech → text → AI cleanup, all in 2–5 seconds
Polished text appears
Auto-pasted where your cursor is — no copy-paste needed
Works in Every App
One global hotkey — works everywhere. No plugins, no extensions, no per-app configuration. Getting it set up properly in Word takes one extra step — see the dedicated guide.
Two Processing Modes
Choose how your voice is processed. Switch between modes anytime — no restart needed.
Local Mode
- 100% offline — voice never leaves your device
- No ongoing costs (included in license)
- 5 model sizes: Tiny to Large
- HIPAA-safe, air-gapped environments
- AI via Ollama (offline, no web search)
App: ~25 MB + model size (75 MB – 3 GB)
Cloud Mode
- Highest accuracy via OpenAI API
- ~$0.003/min (pay OpenAI directly)
- No models to download
- Bring your own API key (zero markup)
- AI via GPT-5 Nano, Mini & GPT-5
App: ~25 MB total — that’s it
Ready to Try It?
Download Whisper, press a hotkey, and start speaking. Free for local transcription, no card required.
Want a detailed setup guide? Check out Local Mode Setup or Cloud Mode Setup.
