Open source · 100% offline

Speak it. It types it.
Nothing leaves your machine.

LocalYapper is fully offline voice dictation for your desktop. Press a key, speak, and your words appear in whatever app you're already in — doc, chat, browser, IDE. No cloud. No subscription. No audio ever uploaded.

Download free See how it works
Free forever MIT licensed Windows · macOS · Linux ~458 MB one-time model
Dictated locally by —+ people — and counting.
The flow

Five steps. All on-device.

Deliberately simple: hotkey to paste, with nothing in between touching the internet.

01

Hotkey

Hold F8 to dictate, or double-tap for hands-free mode.

→
02

Microphone

16 kHz mono audio is captured straight into RAM — never to disk.

→
03

Voice detection

Silero VAD trims the silence so only real speech is processed.

→
04

Recognition

NVIDIA Parakeet transcribes locally. No LLM “cleanup” tax.

→
05

Paste

Text drops into your active app; your clipboard is restored.

When the pipeline finishes, the audio is gone. It is never written to disk.

Built lean on purpose

A no-nonsense tool that gets out of your way.

One job, done privately and instantly. The simplicity is the feature — it's why LocalYapper barely touches your machine.

Truly offline

Parakeet runs fully on-device. After a one-time model download, you can pull the ethernet cable and never look back.

Private by architecture

There is no upload path. No telemetry, no crash reporting, no remote logging. Privacy isn't a policy here — it's the design.

Instant response

A dedicated ~10 KB overlay window stays alive, so recording starts the moment you press — no settings UI in the way.

Works everywhere you type

Clipboard injection means any text field in any app — Word, Slack, your browser, your IDE, email. No plugins required.

Open source (MIT)

Inspect the code, fork it, build it yourself, or self-host. Trust by verification, not by marketing.

Resilient by default

Your words are saved to History before paste is attempted. If an app blocks paste, the text is still right there with a copy button.

The unfair advantage

Half the RAM. Faster. Lighter. $0.

Against cloud dictation tools, LocalYapper's edges are structural — not cosmetic.

½
the RAM
≈ half of Wispr Flow while dictating
1.5×
faster
no cloud round-trip, no LLM stage
30MB
idle footprint
drops from ~400 MB to ~30 MB at rest
$0
forever
no API keys, no tiers, no BYOK
 
LocalYapper
Cloud tools
Where your audio goes
Stays in RAM, on your machine
Uploaded to their servers
Cost
Free, MIT, forever
Subscription / per-use
RAM while dictating
~400 MB, ~30 MB idle
Heavy model resident or cloud
Internet required
No — fully offline
Yes, always
LLM “cleanup” latency tax
None — what it hears is what you get
Extra model, cost & delay
Source code
Open, forkable, self-hostable
Closed

Figures reflect LocalYapper's measured footprint; competitor behaviour is based on their published architectures. Your numbers will vary by machine.

Resource footprint

Built to disappear when you're not using it.

Lazy model loading plus automatic idle eviction means LocalYapper sits near zero until the moment you press the key.

Memory

~30 MB

Idle / cold start

~400 MB

Actively dictating

After ~60 s of inactivity the speech engine is dropped from RAM. The next keypress reloads it in about a second.

CPU

~0%

Idle

5–10%

Capture + transcribe + paste

Native Rust core on Tauri 2 — no Electron bloat. Two separate WebView bundles keep the overlay to the ~10 KB it actually needs.

Available for every desktop

Download LocalYapper

One download, one model install, dictation forever. Pick your platform.

Windows 10+ · .exe (NSIS) Download → macOS 12+ · Intel & Apple Silicon Download → Linux X11 & Wayland · DEB / AppImage Download →

By downloading and installing LocalYapper you agree to our Terms & Conditions and Privacy Policy. LocalYapper is provided “as is”, free and open source under the MIT License, without warranty of any kind. We collect no data — there is no server or endpoint to send it to.

FAQ

Questions, answered.

No. Audio is captured into RAM, transcribed on your machine, and discarded the instant the pipeline finishes. There is no upload path and no endpoint to send it to. The only network requests LocalYapper ever makes are the one-time downloads of the speech and voice-detection models.
It's free forever and open source under the MIT License. No subscription, no API keys, no “bring your own key,” no tiered plans. The catch is that there isn't one — you can read every line of the code and build it yourself.
Only once, to download the ~458 MB speech model on first run. After that, LocalYapper is fully offline — no Wi-Fi, no ethernet, no signal required. Dictate on a plane, in a vault, wherever.
About 30 MB at rest and around 400 MB while actively dictating — roughly half of what cloud tools like Wispr Flow keep resident. The engine is dropped from RAM after ~60 s of inactivity and reloaded in about a second on your next keypress. You can flip a toggle to keep it always-resident if you'd rather trade ~30 MB of idle RAM for instant first-dictation latency.
Any text field in any app. LocalYapper types via clipboard injection, so it works in Word, Slack, your browser, your IDE, email — anywhere you can paste. No plugins or per-app integrations needed.
Windows 10+, macOS 12+ (Intel and Apple Silicon), and Linux on both X11 and Wayland — all first-class, with installers built for every release.
No. There's no LLM tax. What the speech model hears is what gets pasted — fast and predictable. LocalYapper is a dictation tool, not an AI writing assistant, and that focus is the point.
Your words are saved to History before paste is even attempted. If an app blocks the paste, the text is still in your local History and on the overlay with a copy button — you never lose a sentence.