Timed editing
Edit individual segments while retaining valid caption alignment and the original source version.
The product roadmap
A solid core.
A careful next step.
LocalScribe turns existing recordings into useful text on your Mac. The core works; the next priority is making setup and delivery dependable for customers.
Reviewed 13 October 2026 · Current personal build: 0.2.0
Public launch is in preparation. The current build is ad-hoc signed and has been tested on Apple Silicon with macOS 27. Models, ffmpeg and yt-dlp are not bundled. Public checkout stays closed until the release requirements below are met.
Now: the local workspace
The native SwiftUI app supports imports, explicit media-link downloads, local recognition, transcript review and exports. No account or cloud transcript service is required.
Bring recordings into one library
Import audio or video, including multiple files, or inspect and download one media item with yt-dlp. Imported originals are copied into the local library. A sequential queue runs one speech task at a time.
Run a model that supports your task
Local Parakeet and Whisper models are validated before use. Spoken languages depend on each model's capabilities. Compatible multilingual Whisper supports Arabic transcription and a separate English-translation task. Parakeet v3 supports its listed European languages and does not translate.
Review while keeping the source
Listen to recordings, including audio from video files, change playback speed and seek using timestamps. New recognition saves a new version. Editing creates a separate copy while preserving the original recording and earlier output. Supported right-to-left scripts have appropriate reading layouts.
Export useful formats
Export TXT, Markdown, JSON, PDF, SRT or VTT. Timed captions need timing from recognition; free-text edited copies have no caption alignment. Text and document exports preserve Unicode.
Keep processing local
Recognition, model checks and audio conversion run in local processes with network access denied. Inspecting links and explicitly downloading media or models use the internet. Local storage has owner-only permissions, rather than app-level encryption.
Existing verification includes 26 tests, local English and Arabic recognition, French-to-English translation and direct-media downloading. This does not establish support for every media website, model or Mac.
Next: a dependable first release
Before we accept payment, a customer should be able to install the app and complete a first transcript without an existing development setup.
Sign and notarize the app
Use Developer ID signing for the app and helpers, complete notarization, and verify Gatekeeper accepts the downloaded installer.
Test setup on clean Macs
Validate the advertised Mac and macOS combinations. Make tools and models easy to locate or install, with clear sizes, sources and readiness errors.
Make purchase and delivery reliable
Confirm the seller's eligibility and payment route. Test receipt and download delivery, and publish the customer license, refund policy, update entitlement and support scope.
Preserve third-party notices
Review dependency and model terms for the chosen distribution. Include the required notices and explain which tools and models are separate downloads.
Further priorities include library backup and deletion, keyboard and VoiceOver review, measured performance guidance, and updates that preserve existing recordings and versions. Any remaining limitations will be disclosed when checkout opens.
Later: better transcript review
These are proposed improvements, ordered around the work people already do with transcripts.
Version comparison
See changes between recognition, translation and edited copies, with each version's model, task and language retained.
Recording organization
Explore folders, tags and clearer batch controls without losing recordings, jobs or versions.
Tested link coverage
Check specific sources reproducibly and make source errors easier to understand. Claims will follow actual testing.
Ideas to evaluate
Live microphone recording, speaker labels, local summaries or chapters, and Windows or Intel support are research candidates. They are not implemented or promised.
Each needs its own validation: reliable recording permissions, speaker accuracy, source-linked summary review, or a separate platform support and performance matrix. Priorities may change after feedback.
The launch model
The proposed price is $10 USD at launch, then $15 USD regularly, as a one-time purchase. The final included version, updates, support and refund terms will be published before checkout opens. We have not committed to lifetime updates or unlimited support.
Start with a small paid group after the release checks pass. Use voluntary feedback on setup, first transcripts, exports and support needs to decide what to improve. The app currently has no analytics.