Multilingual Whisper
Transcribe supported languages, including Arabic. Compatible multilingual models can also produce a separate English translation.
English-only, Turbo and distilled variants are not offered for English translation.
Local transcription for Mac
Your audio.
Your words.
Turn recordings into readable transcripts on your Mac, with local models and a workspace that keeps the source.
From a recording
to a useful page.
One workspace for the whole journey, with the original always within reach.
Bring it in
Import audio or video. Or inspect a media link and choose to download it.
Find the words
Choose a compatible local model and spoken language. Start transcription when you are ready.
Listen. Refine.
Jump back to the audio with timestamps. Save edits as separate versions, keeping the original output.
Take it with you
Export a readable document, structured transcript, or timed captions when alignment is available.
A local first approach
The work happens
on your Mac.
Your speech model runs locally. Recognition and audio conversion run in separate processes with network access denied.
Read the privacy detailsYour library, on disk
Recordings, transcript versions and job history are stored locally with owner-only file permissions. This is not an encryption feature.
Downloads are deliberate
Inspecting media links, downloading media and downloading models use the internet. Recognition does not use a cloud service.
A small, focused app
No account, app analytics, remote transcript service or cloud AI. Your original recording and earlier transcript versions are retained.
Choose the model.
Keep the language.
LocalScribe checks a model's capabilities before a task begins. Language support depends on the model you choose.
Parakeet v3
Transcribe its supported European languages locally. Parakeet does not support Arabic and does not translate into English.
The app helps you switch to a compatible model while keeping your selected language.
Models are not bundled. Use a compatible model already on your Mac, validate a model file, or explicitly download one. Arabic and other supported right-to-left scripts have reading layouts and Unicode exports.
Words that travel.
Six ways to move your transcript into the next part of your work.
TXTMDPDFJSONSRTVTT
SRT and VTT need timing from the speech model. Free-text edits have no caption alignment, so edited copies export as text, Markdown, JSON or PDF.
Launch in preparation
A useful tool.
A simple purchase.
We are planning a one-time price for LocalScribe, with a smaller launch price for early customers.
Checkout is not open yet. Signing, notarization, clean-Mac setup and distribution testing come first. Final pricing and purchase terms will be published before sales begin.
Planned one-time launch price
$10USD
Planned regular price $15
The speech models are separate downloads. A public app download will follow distribution readiness.
Ask about launchOpens your email app. No mailing list or account.
Built with care.
Growing with purpose.
The core is working. The next priority is making it ready for someone else's Mac.
The local workspace
Import and download media, run local recognition, preserve versions, review with audio, and export in six formats.
A dependable release
Developer signing and notarization, clearer first-run setup, clean-Mac testing, model guidance, and purchase support.
Explore what helps
Timed editing, version comparison, recording organization and tested link coverage are proposed improvements.
Priorities may change. Later ideas are not release promises.
Read the full roadmapA few useful answers.
Can I buy or download LocalScribe today?
Public checkout and distribution are not open. The current app is a personal, ad-hoc signed build. We are preparing signing, notarization, setup instructions and testing before a public sale.
Which Macs are supported?
The current build has been tested on Apple Silicon with macOS 27. Older macOS versions and Intel Macs have not been validated. Broader support will be assessed as part of the roadmap.
Does it work without the internet?
With the required tools and a compatible model available locally, recognition and audio conversion run without network access. Checking links and downloading media or models require the internet. ffmpeg and yt-dlp are currently installed separately.
Can it transcribe Arabic or translate into English?
Arabic transcription requires a compatible multilingual Whisper model. Parakeet v3 does not support Arabic. English translation is a separate task available only with compatible multilingual Whisper models. Language options depend on validated model capabilities.
Will my recording or original transcript be overwritten?
Imported originals are copied into the local library. Recognition saves a new version, and editing creates a separate copy. Previous versions remain available. Review recognition output before relying on it.
Does it include live recording or speaker labels?
It currently works with imported or downloaded media. Live microphone recording, speaker diarization, summaries, license activation and an automatic updater are not implemented.