Get started with SpeakLog on Mac

1. Install and check the requirements

The current public release is 0.1.1 and requires an Apple silicon Mac (M-series chip) with macOS 26 or later. This package does not support Intel Macs. Download the PKG and follow the installer. The release page also offers a DMG and SHA256SUMS checksums.

Transcribing a language for the first time may require a system speech-model download. Available transcription languages depend on system support detected by the app. Choosing a practice language does not guarantee offline recognition for that language.

2. Review language and AI settings first

Set your practice language, explanation language and learning level. The explanation language is used for correction reasons, translations and vocabulary definitions.

Automatic AI review after import is on by default. Turn it off in Settings → AI before importing if needed. Before using private material, check automatic review and provider settings; disable automatic review if you only want a transcript first. Cloud AI sends its input to the selected provider.

Claude Code is the default provider. Install and sign in to Claude Code or Codex before choosing its CLI provider; use the test button in Settings → AI to check the connection. The Anthropic API uses your own API key. These providers and the on-device model have different requirements. The local model requires enabled Apple Intelligence and an available model. Visual analysis has a separate CLI route: choosing a local model globally does not make every feature offline.

3. Import a recording or video

Choose files from the home import area or drag them into the window. Selecting multiple files creates one project with separate results for each item. You can also record audio in the app; record video elsewhere and import it.

Imported media is normally referenced at its existing location. Keep those files and their drives available. New in-app recordings are saved with the project. Try a short clip first to check the language and audio quality.

4. Review the text, then choose an output

Listen to the original and fix misheard names, words and sentence boundaries. Then compare the natural wording and its explanations. AI suggestions can be wrong; check that they preserve your meaning.

Subtitles, blog posts and videos can be made independently. Exporting is not publishing: you copy or save the result and upload it to your chosen platform yourself.

If something does not work

For transcription issues, check system language support, speech-model availability, the original file location and recording permissions. For AI issues, use the provider test in settings first.

Speed varies with the chip, clip length, language and model state. A single development test is not a general performance guarantee.