OnWhispr user guide

1. Requirements

An Apple silicon Mac (M1 or newer) running macOS 14.0 or later.

Allow about 5 GB for the recommended model combination. Whisper Small is the smaller speech option at roughly 464 MB; the formatting model is optional.

2. First launch

Open OnWhispr and follow the four-step setup. Start with microphone access, then enable Accessibility if you want direct insertion into other apps.

Choose and download a speech model. Add the optional local formatting model for email drafts, structured notes, meeting notes, and custom prompts.

Finish with a real first dictation. The setup only completes after the core path is ready, so missing permissions and models remain visible instead of being silently skipped.

3. Dictate anywhere

Place the cursor in a text field. Hold the configured shortcut (Control-Option-Space by default), speak, and release it. OnWhispr transcribes locally and applies the selected mode.

The Mac App Store edition always copies the result for you to paste with Command-V. The direct edition can optionally insert it through Accessibility and falls back to the clipboard. Your text remains recoverable if processing or delivery fails.

4. Choose the right output

• Raw transcript — your words without formatting; no LLM required.
• Clean dictation — lightweight local cleanup without an LLM, or richer formatting when the optional LLM is enabled.
• Structured notes — summary, bullets, actions, and open questions.
• Email draft — a clear email structure from spoken intent.
• Meeting notes — decisions, risks, and follow-ups.
• Custom prompt — your own local formatting instructions.

5. Microphones and permissions

Select the input device under Settings › Audio. The choice is specific to OnWhispr; the app does not replace your Mac's system-wide default microphone.

If a device disappears, reconnect it and reopen the microphone menu. If macOS shows access as denied, open System Settings › Privacy & Security › Microphone and enable OnWhispr.

6. Models and offline use

The recommended speech model is Whisper Large v3 Turbo (about 3 GB). The recommended formatting model is Qwen2.5 3B Instruct (Q4) (about 2 GB).

Downloads are verified before installation. Compatible local GGUF formatting models can also be linked manually with a valid SHA-256 checksum. After installation, the core dictation flow works without internet.

7. History and privacy

History is optional and stores up to 30 entries locally. It is not separately encrypted at application level. Disable history for sensitive work or clear it from Settings whenever needed.

OnWhispr has no account or analytics SDK and does not upload dictated audio or text to an OnWhispr server.

8. Troubleshooting

• No audio level: verify the selected device and microphone permission.
• Shortcut does nothing: choose a different combination; the shortcut itself does not require Accessibility or Input Monitoring.
• Model unavailable: open Settings › Models and retry or select an installed model.
• Text was not inserted: paste from the clipboard or retry the pending dictation shown by OnWhispr.
• Still stuck: export diagnostics from Advanced Settings and email hello@onwhispr.com. Review the file before sending it.