Detailed notes for every public Direct Mac release.
Features, fixes, and technical changes are grouped by version. Older builds remain historical records; download the current signed release unless support asks you to use another build.
Detailed build notes are currently available in English.
Current release
Diktator 2.3.1 · Build 88
Edit results and choose corrections. Set separate modes for dictation and proofreading. Restore an existing website purchase by email and optionally sync your dictionary with private iCloud.
Editing, individual corrections, separate recognition modes, the personal dictionary, filler removal and iCloud sync require Pro or an active trial. Free includes Small-model dictation and Copy with no expiry. Start the 21-day trial only when you choose.
Edit a result
Edit your dictation before inserting or copying it. Compare the original and proofread text, with red deletions and green additions, and accept or reject each change separately. Direct can insert the edited draft when its original destination is available. In the Mac App Store edition, copy the edited draft and paste it into your app; ordinary dictation continues to use macOS Services.
Recognition mode · Personal dictionary
Choose Faster, Balanced or Thorough independently for dictation and proofreading. Save names, preferred spellings and replacements in your personal dictionary; add entries from the editor’s context menu. Use an editable filler-word list with optional removal, off by default. Preferred spellings can provide optional recognition hints. They do not train Whisper or guarantee a spelling.
Dictionary storage
On-device storage is the default; private iCloud sync is optional. Save to merge your dictionary, filler words and their options with your private iCloud. Audio and transcripts stay on this device. Turning sync off keeps both copies. Dictionary sync requires Pro or an active trial. Sync works between Diktator versions that support this iCloud dictionary format.
Restore access
Use the email associated with your Diktator purchase. Choose “Existing website purchase”, then “Send verification code”. Enter the emailed 8-digit code and choose “Verify purchase”. Email restoration is optional. License-key activation and existing Direct licenses remain available. For an App Store purchase, choose “Restore with Apple Account”. Email restoration is an optional way to verify an existing website purchase; it does not replace Apple purchase restoration. Direct and App Store installations each use one device slot. Your email is sent to Diktator to verify this purchase. Recordings and dictated text are not sent.
Help
Updated onboarding, model descriptions and Help explain these features in all 13 interface languages.
These notes cover Direct Mac 2.3.1 and the prepared Mac App Store adaptation. Sync works only between Diktator versions supporting the same iCloud dictionary format. Whisper model files and the recognition engine are unchanged.
The model files, translation options, and licensing are unchanged. Apple silicon Macs running macOS 26 or later remain supported, including macOS 27.
Reduced ancillary network requests in the Mac edition from the website and replaced the computer name with a neutral device label in license requests. Clarified on-device Apple translation and privacy information.
The model files, translation options, and licensing are unchanged. Apple silicon Macs running macOS 26 or later remain supported, including macOS 27. The transcript window, advanced audio tuning, and final proofreading remain Pro features.
Removed the general English hint about keyboard names that was sent with every dictation. Recognition no longer receives that unrelated English context. Diktator no longer replaces spoken punctuation names with symbols in its own text processing. The speech model continues to add punctuation naturally. When enabled, the transcript window now shows recording and processing status, then the final text, even without segmentation. It does not add another recognition pass. When enabled, the noise threshold and stop-on-silence work with or without segmentation. You can choose how to record without losing these controls. Settings panels keep their width and alignment when options change or the window is resized. The dictation instructions in Settings and first-run setup are more compact. Optional final proofreading applies to segmented recordings. Without segmentation, the recording is recognized once after Stop, without a redundant proofreading pass. The model files, translation options, and licensing are unchanged. Apple silicon Macs running macOS 26 or later remain supported, including macOS 27. The transcript window, advanced audio tuning, and final proofreading remain Pro features.
Updated for macOS 27, with more reliable text insertion and a recovery window that keeps your result available.
Build 77 verifies text and cursor position after Accessibility insertion, prefers the keyboard helper in Codex when enabled, and avoids retrying an uncertain insertion through another method. When insertion or proofread replacement is refused, the full result remains in the recovery window with Copy All. Built with the macOS 27 SDK; Apple silicon and macOS 26 or later are still required. Speech models, translation, Free/Pro access, and Direct licensing are unchanged.
Build 76 moves the Direct edition onto the shared Diktator processing foundation while preserving its website licensing and delivery features.
Build 76 improves short-recording evaluation, isolates translation requests so late results cannot replace newer output, preserves retained text for Apple Translation after a proofreading failure, aligns clipboard delivery with the selected settings, and keeps a saved larger-model choice during fallback.
Build 75 changed the version and package metadata only. Dictation, selectable translation, text insertion, model handling, licensing and all 13 interface languages retained the stable Build 70 behavior.
Translate speech from the broad supported source-language catalog into a selected target language. Availability and quality depend on the language pair, model, and translation method. Local-model translation is experimental; review important results.
Text insertion and setup
Improved input-method helper setup: after installation and macOS approval, switching the option off and on reuses the approved component.
Improved focus handling after text insertion and prevented stale helper activation callbacks from taking focus from the writing app.
The website edition uses its own dictation shortcuts and insertion routes; it no longer registers a legacy macOS Service.
Refined onboarding, translation settings, processing choices and status colors.
Free mode and licenses
Improved text insertion and model handling after returning to the free version. The free version uses Small; activating the advanced version restores the previously selected model when available.
License summaries display the actual Mac limit returned by the licensing service. An unavailable limit is left unknown instead of assuming three Macs.
Clearer localized activation errors distinguish invalid or inactive keys from connection and service failures.
Existing local dictation, model translation, Apple Translation and the 13 interface languages remain available according to the selected edition.
Known issue
Command-V may not paste a license key in the activation dialog. Right-click the field and choose Paste from the context menu. This keyboard-shortcut issue is deferred to a future release.