Speak naturally. Your Mac writes at the speed of thought.|
Hold ⌥ Option and speak freely. Local AI transcribes and cleans speech hesitations instantly, typing directly wherever your cursor blinks. 100% Offline. 100% Open Source. Zero accounts. Zero latency.
# Caspr Architecture — Local Dictation
- Ultra-fast inference and instant typing
|
Why standard dictation is frustrating,
and how Caspr reinvents it.
Classic dictation engines write every 'um', repetition, and stutter. Caspr understands your train of thought and delivers clean, structured, and punctuated text.
Raw & Cluttered with Hesitations
"So um wait, let's create the function um get user data no wait fetch user data with a try catch comma and um if it fails return null question mark"
- ⚠️ No removal of stutters or false starts
- ⚠️ Awkward oral punctuation you have to say out loud
- ⚠️ Catastrophic handling of code terms and technical jargon
Cleaned, Structured & Punctuated
"Let's create the fetchUserData function with a try/catch block, and return null on error."
- ✓ Automatic removal of 'ums', repetitions and false starts
- ✓ Natural punctuation and capitalization inferred from sentence context
- ✓ Flawless recognition of code variables, acronyms and tech terms
Two Engines Included.
Tailor Caspr to your Workflow.
Switch in 1 click depending on your hardware specs or precision needs.
macOS Native Engine
Leverages Apple's built-in Speech recognition frameworks directly. Zero extra dependencies, instant startup and near-zero RAM footprint.
- Setup: Instant (0 MB download)
- Speed: Ultra-fast (< 0.2s)
- Best for: Quick notes, lightweight typing on any Mac
CrisperWhisper AI Engine
Advanced local neural transcription model optimized for Apple Silicon. Eliminates stutters and structures sentences with pinpoint accuracy.
- Model: Fast local quantized neural network
- Technical terms: Flawless code & mixed languages
- Best for: Developers, writers, professional emails & meetings
A macOS app engineered for raw speed.
Pure native Swift engineering combined with local AI models optimized for Neural Engine and GPU.
100% Local. Zero Cloud. Zero Data Leaks.
Your voice and text never leave your Mac. Inference runs locally on Apple Silicon (Metal/MPS) or Intel CPU. Turn off your Wi-Fi: Caspr runs exactly the same.
⌥ Option Key or Custom Shortcut
Hold the Option key (or your custom shortcut), speak your thought, and release: text is typed instantly. No clicking, no switching windows.
Zero focus stealing
A lightweight non-activating panel (NSPanel): it displays your live audio waves and streaming text without ever unfocusing the app you are typing in.
At Cursor or Append to Notes
Dictate directly into VS Code, Slack, Notion, Word... Or continuously append your thoughts to a daily Markdown journal without interrupting your task.
Metal GPU & Neural Engine
Ultra-fast inference optimized for Apple Silicon M1, M2, M3, M4 and Intel processors with minimal battery impact.
24+ Languages & Hotwords Dictionary
Native support for 24+ languages with instant switching. Add your custom acronyms, variable names, and technical terms to your custom dictionary for pinpoint accuracy without manual edits.
100% Open Source. Free. Auditable.
Caspr belongs to the community. No hidden subscriptions, no paywalled features, no opaque proprietary code.
Public Source Code on GitHub
All Swift code (AppKit, SwiftUI) and the Python engine (uv, PyTorch, Whisper) are open source. Inspect every single line, fork the repo, or build it yourself.
Forkable & Modular
Train custom models, tweak transcription prompts, or integrate custom audio pipelines to fit your exact developer workflow.
Clean Uninstaller & Zero Residue
Caspr includes a dedicated clean uninstaller that completely wipes the Python environment, caches, and models without leaving orphan files on your Mac.
Everything you need to know about Caspr
Straight answers to common questions about privacy, offline inference, and setup.
Caspr is compatible with macOS 14 (Sonoma), macOS 15 (Sequoia) and later. It runs at full speed on Apple Silicon chips (M1, M2, M3, M4) using Metal/MPS GPU acceleration, and fully supports Intel Macs.
No, never. Caspr operates completely offline. The app has no backend servers, no user accounts, no cloud API connections, and zero telemetry. Your voice strictly remains in your Mac's memory.
To distribute software without Gatekeeper prompts, Apple requires a recurring $99/year developer subscription. As a 100% free and open-source project, Caspr is distributed directly. Following the 1-click step (Open Anyway in System Settings) permanently authorizes the app on your Mac.
It's Hold-to-talk: simply hold the Option key down while you speak, then release it. The moment you release, Caspr cleans the text and instantly types it right where your cursor is.
Yes, absolutely! The AI model seamlessly understands technical jargon, code snippets (async/await, Pull Request, database) and you can add your custom hotwords dictionary for personalized terms.
Ready to write faster than you can type?
Download Caspr for free and experience the power of next-generation local voice dictation on your Mac.
