Hex App Guide: Free Local-First Voice Dictation for Desktop
Turn speech into text anywhere on macOS, Windows, and Linux with Hex, an open-source local-first dictation app. Covers setup, modes, labs, and privacy.
Table of Contents
- 1. What Is Hex?
- 2. How It Works
- 3. How to Install Hex
- 4. One-Time First-Run Setup
- 5. Everyday Controls
- 6. Interactive Settings Explorer
- 7. Standout Features & Labs
- 8. Is My Voice Private?
- 9. Frequently Asked Questions & Troubleshooting (FAQ)
- 10. Community & Quick Glossary
1. What Is Hex?
Hex turns your voice into typed text directly wherever your cursor is—notes, chat apps, search bars, documents, or code editors—across macOS, Windows, and Linux.
Four Core Promises
- Zero Friction — Hold your shortcut while speaking, release when done. No save button required.
- Runs Locally on Your Machine — Speech-to-text happens directly on your computer; your voice never leaves your device.
- Pastes Anywhere — Works in any text field without per-app plugins.
- MIT Open Source — Inspect the code, learn from it, or fork it freely.
2. How It Works
Dictating with Hex boils down to three simple motions: Hold → Speak → Release.
- 1Hold your shortcut
A compact floating HUD appears on screen so you know Hex is actively listening.
- 2Speak clearly
Talk at a natural pace while the 6-dot scope reacts live to your voice volume.
- 3Release the key
Moments later, your computer transcribes the clip locally and pastes the text right at your cursor.
Modern desktops handle speech recognition well on their own. Once you download a speech model on first launch, Hex works completely offline.
3. How to Install Hex
Download the installer for your OS from anoma.ly/hex or the README's Downloads table. New users should pick Stable (while Dev previews early features that may occasionally break).
| Platform | Download Package |
|---|---|
| macOS (Apple Silicon: M1–M4) | .dmg installer |
| Windows (x64 PC) | Installer |
| Windows (ARM64 laptop) | ARM64 Installer |
| Linux (Debian · Ubuntu) | .deb package |
| Linux (Fedora · openSUSE) | .rpm package |
| Linux (Other distros) | Standalone Executable |
Install via Package Manager / CLI
brew install --cask anomalyco/tap/hex4. One-Time First-Run Setup
- 1Allow Microphone Access
Click Allow when prompted so Hex can hear your voice.
- 2Allow Text Insertion
Enable Accessibility permission on macOS (Windows requires no extra step; Linux guides you based on your desktop environment).
- 3Download Your Speech Model (Once)
Requires internet once and may take a few minutes depending on model size. Afterward, dictation works fully offline.
5. Everyday Controls
Hold ModeDefault — record while holding the key
“Hold → Speak → Release”
- Behavior
- Records while pressed; transcribes and pastes upon release
- Mistap Guard
- Ultra-brief accidental taps are automatically ignored
- Best For
- Chat messages, quick search terms, short comments
Toggle ModePress once to start, press again to stop
“Hands-free long-form dictation”
- Behavior
- Enable under General > Shortcut mode to start/stop with single presses
- Cancel Key
- Press Esc while recording to discard with zero trace in history
- Best For
- Emails, documentation, long prompts
Lock ModeDouble-tap while holding to lock recording on
“Release your fingers and keep thinking aloud”
- Behavior
- While holding, tap twice lightly to lock recording hands-free
- Benefit
- Best of both Hold and Toggle without changing settings
- Best For
- Spontaneous longer explanations and outlines
Three Everyday Safeguards
- Safe Cancel (
Esc): Cancels active recording only (won't abort already-finished transcription), and never hijacksEscin other apps when you aren't recording. - Paste Last Dictation: Re-pastes your most recent transcript if a paste missed or you want it in another window.
- Silence Detection: Holding the key without speaking produces nothing—preventing phantom hallucinations from quiet background air.
6. Interactive Settings Explorer
Hex splits its interface into a left sidebar (General · Output · Modes · History · Labs) and a right panel. Every toggle saves and applies immediately—there is no Save button.
- Configure shortcut key, Hold/Toggle mode & Cancel key· General
Click the key box and press your desired combo. Selecting Toggle reveals the Cancel key setting (defaults to Esc).
UI Label:Shortcut / Shortcut mode / CancelSection:General - Pick speech model, spoken language & custom vocabulary· General
Choosing your language recommends matching models; Vocabulary lets you register names and jargon for higher accuracy.
UI Label:Speech model / Language / VocabularySection:General - Microphone source, auto-launch & chime volume/theme· General
Choose System default mic, toggle 'Turn off the mic when not dictating', auto-start at login, and preview wood/keyboard/bell sound packs.
UI Label:Microphone / Launch at login / SoundsSection:General - Fix Bluetooth mic drop, pick GPU/CPU engine & keyboard sync· Advanced
Use Mac's built-in mic when wearing Bluetooth headphones, choose Auto/GPU/CPU, auto-switch models with keyboard language, or bind left/right modifier keys.
UI Label:Use Mac's mic / Speech engine / Switch modelSection:Advanced - Duck other audio, move HUD & enable terminal paste· Advanced
Keep, mute, or lower other audio (Lower to %), position HUD (Top/Bottom/Off), enable Ctrl+Shift+V terminal paste on Linux/Windows, and restore GNOME clipboard.
UI Label:While dictating / Dictation HUD / Terminal pasteSection:Advanced - Dock icon, interface language & permission shortcuts· Advanced
Toggle Dock icon visibility, switch UI language, jump to OS permissions, and bind Paste last dictation.
UI Label:Show Dock icon / Interface language / Open settingsSection:Advanced - View processing pipeline & set raw-paste bypass shortcut· Output
Shows numbered post-processing steps top to bottom (when empty, Hex pastes your exact words). Bind Dictate without processing to skip all steps on demand.
UI Label:Every dictation / Without processingSection:Output - Test output with Try or hand rules to an AI agent· Output
Preview What Hex pastes via Try, copy an agent-ready prompt via Edit with agent, or edit the config directly via Open file (network/command rules require Allow once).
UI Label:Try / Edit with agent / Open file / AllowSection:Output - Auto-switch rules per frontmost app (Chat, Terminal, etc.)· Modes
Global acts as the fallback; app-specific modes automatically take precedence whenever that app is active.
UI Label:Global / + / Applications / Switch when frontmostSection:Modes - Phrase corrections & optional AI rewriting· Modes
Map spoken phrases (e.g., 'open code' → 'OpenCode') or opt in to AI rewriting with custom Instructions and Deadline seconds.
UI Label:Corrections / OpenCode transformationSection:Modes - Search history, retention window & dictation stats· History
Set retention (Off / 24 hours / 7 days / 30 days / Forever), inspect app, model, latency, and raw transcript, or view usage totals under Stats.
UI Label:Search history / Kept 7 days / Stats / CopySection:History - Keep audio clips & re-transcribe with another model· History
With audio retention enabled (Lab), replay clips, run Transcribe again with a stronger model, and swap in Earlier versions.
UI Label:Keep audio too / Play / Transcribe again / ReplaceSection:History - Explore opt-in experiments or re-run initial setup· Labs
Toggle experimental cards grouped by Dictation, Meetings, Annotate, Models, etc. Use Run setup again anytime permissions get tangled.
UI Label:Discover / Turned on · N / Search labs / Run setup againSection:Labs
7. Standout Features & Labs
Meetings (Beta, macOS first)Meetings · VisualsCaptures online meeting audio, signals start and end, and saves a full transcript automatically once the call wraps.
AnnotateMeetings · VisualsSnap your screen and draw circles or highlights while speaking—handy for walkthroughs and bug reports.
Custom VocabularyAccuracy · HistoryPre-register names, schools, brands, and technical jargon so Hex recognizes them accurately every time.
Searchable History & Re-TranscriptionAccuracy · HistoryRecover lost paragraphs effortlessly and re-run saved audio through newer speech models.
CLI Control (hex dictation start)CLI · LabsTrigger dictation from terminal scripts, automation tools, or AI agents directly.
Hidden Labs GemsCLI · LabsOpt into live text preview while speaking, spoken punctuation (commas, line breaks), filler-word removal ('um', 'uh'), and per-app auto-pause.
8. Is My Voice Private?
There is one explicit opt-in exception: if you enable AI rewriting (OpenCode transformation) inside a Mode, the transcribed text is sent to the configured AI model to polish. It is off by default, so nothing is sent unless you consciously turn it on.
9. Frequently Asked Questions & Troubleshooting (FAQ)
Q1. Is Hex completely free, and does it work offline?
Q2. Does Hex support multiple languages and Windows/Linux?
Q3. Does my voice or transcript get sent to external servers?
Q4. Pressing my shortcut does nothing.
Q5. I'm speaking, but no text appears or it won't paste into other apps.
Q6. Words come out garbled, or the very first dictation feels slow.
Q7. How do I update Hex or report bugs?
10. Community & Quick Glossary
Built by Anomaly in Rust (with UI rendered by Rune), Hex has earned ~6,400 GitHub stars and 400+ forks as of November 13, 2026, complete with transparent docs/ covering features, design plans, and release notes.
| Term | Plain-English Meaning |
|---|---|
| Dictation | Converting spoken words into typed text automatically |
| Speech Model | Pre-trained local "brain file" downloaded once to understand speech |
| Local-First | Processed directly on your own computer rather than a remote cloud server |
| HUD | Floating mini overlay showing active recording status |
| Shortcut | Key combination that triggers dictation in one motion |
| Clipboard | Temporary OS memory holding copied text before pasting |
| Toggle | Mode where each key press switches recording on or off |
| MIT License | Permissive open-source license allowing free use, modification, and sharing |
| Beta | Nearly complete feature currently being tested in the wild |
| Labs | Opt-in area for early experimental features |
Polish Dictated Drafts Locally with Hafuture Spell Checker
Proofread Korean, English, Japanese, or Spanish drafts you dictated with Hex—100% locally in your browser with zero file uploads.