User guide

Quick start

1. Open BlabToText from Applications. Find the BlabToText voice-bars-and-text mark in your Mac's menu bar.
2. Complete setup with On-device only. Allow Microphone and Input Monitoring when asked; Apple Speech may also request Speech Recognition permission or a language download.
3. Check your saved Record Shortcut Key. Hold that key, say a short sentence, then release. The factory default is Right Option; if you chose Right Control, use Right Control.
4. Wait for transcription. The default is to copy the result. Click a text field and press ⌘V.

Use Settings → Setup Assistant to revisit setup. Apple on-device availability depends on your Mac, macOS version, and language. English is the initial focus; other languages are not yet broadly validated.

Opening the next beta: first use offers setup. After setup is saved, opening BlabToText shows a readiness window with your actual shortcut, recording mode and permission status. Closing it leaves the menu-bar app running. Choose Ready to Record to reopen it, or Latest Transcript to review text. Readiness does not test a cloud connection or guarantee speech availability. TestFlight's separate From the Developer page contains beta test notes, not BlabToText settings.

Two ways to finish a thought

Open Settings → Recording Mode from the BlabToText menu.

• Hold to talk: hold your shortcut while speaking; release to finish. This is the default.
• Tap once, finish after silence: tap and release to start, then speak. A pause ends that recording. Tap again to finish early.

Set Finish after silence between 0.5 and 10 seconds (default: 2). Choose a sensitivity preset for your voice and room. Detection uses microphone volume, so background conversation or steady noise can delay completion.

The countdown ends recording, not transcription. Processing time follows before copy or paste. The app does not stay listening after each dictation.

Click the indicator's ×, press Escape, or use the menu's Cancel command to cancel recording or transcription. No sustained input for 15 seconds cancels pause-mode recording. Dictation has a safety limit of 120 seconds, or 55 seconds when the older on-device recognizer may be used.

The next beta's menu-bar icon uses system-adaptive contrast, with different shapes for recording, processing, copied and error states. The menu and tooltip explain the current state; color alone is not the status signal.

Your text. Your control.

Copy: available in every Mac edition and enabled by default. Paste with ⌘V when you are ready. The menu's delivery line shows the effective setting.

Automatic paste (direct Mac build): open Settings → Recording Mode, enable automatic paste, and grant Accessibility permission. Place the cursor in an editable field before starting. Keep that same app, window, and field focused until transcription finishes. If BlabToText cannot verify the target, it copies the text instead. Some apps do not expose enough accessibility information for automatic paste.

The TestFlight/App Store build shows automatic paste as unavailable. Granting Accessibility does not add this missing feature; automatic stopping and copying still work.

Enter after a pause (direct Mac build): with automatic paste and pause mode enabled, turn on Press Enter after pause and automatic paste in Recording settings. It is off by default and inactive with copy-only delivery or hold-to-talk.

Some web editors expose only app/window identity, so BlabToText cannot detect every cursor change within them or prove that a paste was accepted. Leave Enter off when you want to review before sending.

If paste does not happen, use ⌘V or open Latest Transcript. Do not repeat a dictation before checking whether the text is already on your clipboard.

A place to review your words

Choose Latest Transcript from the menu. Edit the text, Copy it, Revert to its original wording, or Clear the workspace. When a new result arrives while you are editing, your draft stays intact; choose Load Latest when ready to replace it.

This workspace is held in memory for the current session. It is not a saved transcript history. Copy anything you need before quitting or restarting BlabToText. New dictation does not automatically open this window or steal focus.

Transcribe an audio file on your Mac

1. Open Latest Transcript → Import Audio.
2. Choose an audio file supported by macOS, up to 90 minutes long.
3. Watch progress as Apple on-device speech transcribes short sections.
4. Review and Copy the result when ready. If you were editing a draft, choose Load Latest to see the import.

Imports always use Apple on-device speech, even when dictation uses a cloud provider. There is no cloud fallback for import. Timestamp headings identify audio sections, not individual words. Cancel keeps completed sections; the original audio file is not changed.

Imported results are not automatically copied, pasted, or sent. Speaker identification, meeting joining, calendar scheduling, and meeting summaries are not part of BlabToText.

Private by default. Cloud by choice.

Open Settings → Transcript Agent to choose on-device only, cloud only, or on-device-first routing with an optional cloud fallback. New installs keep cloud fallback off. Local speech recognition can require an initial Apple language download.

For cloud dictation, choose a cloud mode in that same dialog, select OpenAI, Groq, xAI, or Google, and enter that provider's API key. Each provider has its own Keychain entry. In the next beta, switching providers keeps each unsaved key, model choice and completed check with its provider while the dialog is open. Save lists all entered keys to store separately and activates the selected mode, provider and model. Cancel discards drafts. Saving in on-device mode does not apply unsaved cloud edits. xAI/Grok is not Z.ai/GLM; Z.ai is not supported in this beta.

Keys are stored in Apple Keychain. Use Remove Key in the same dialog to remove a selected provider's credential. Requests go directly to your chosen provider; provider billing and retention policies apply.

Bring your own key (BYOK): use an API key from your provider’s developer account. Provider API usage is billed separately from ChatGPT or Claude subscriptions. BlabToText does not connect to or reuse credentials from those desktop apps. Claude is not a transcription provider in this beta. On-device only requires neither an account nor a key.

Connection check (next beta, not installed build 2): provider selection and key entry share one dialog. The named button, such as Test xAI Connection, sends the selected provider’s key to its official model-list API, without audio. Matching supported audio models appear in the list; a current unlisted selection is kept and labelled. A successful list request does not prove transcription permission, billing or quota—verify those with a separate, explicitly chosen dictation test. Testing neither saves the key nor enables cloud routing.

The official host is shown next to Test; expand Connection details for exact endpoints and the advanced manual model ID option. Official provider endpoints are built in; there is no URL to enter for these providers. The model catalog cannot change destinations or add a new protocol. New providers and endpoint changes require an app update. Custom endpoints are not supported by this connection-check flow.

Where the models come from: before testing, the dropdown shows supported suggestions bundled with BlabToText, not models verified for your key. After testing, it shows compatible audio models returned by the selected provider. xAI chooses its speech model automatically; its general model list does not validate speech access. Supported suggestions and endpoints change through app updates. The next beta removes build 2's confusing Refresh Providers & Models command and stops contacting its unavailable legacy catalog host. Read the privacy policy for details.

Make it yours

• Record Shortcut Key: choose Right/Left Option, Right/Left Control, or Fn. If Right Option is used for accented characters on your keyboard, choose another key.
• Custom Vocabulary: add names or specialized terms and common mishearings, one entry per line. For example, BlabToText = talk to me. Local corrections apply to dictation; supported cloud requests also receive vocabulary hints.
• Play Sound Cues: enable or disable audible feedback.
• Launch at Login: start BlabToText when you sign in to your Mac.

Permission shortcuts and a three-second microphone test live under Help. The test transcribes and copies a result; it never automatically pastes or presses Enter. You never need to open a settings folder or edit a file.

Know your Mac edition

Both target macOS 13 and later. Apple on-device support varies by device, OS, and language. Public distribution is still being prepared; this table describes the builds, not App Store availability.