Handy vs OpenWhispr vs Voconly for Offline Dictation

Handy is a good first trial for a shortcut that puts a local transcript in your editor. OpenWhispr suits people who want local dictation and the option of hosted transcription, cleanup, or meeting features in the same app. Voconly suits Windows x64 and Apple Silicon users who want to set up distinct shortcuts for different cleanup tasks. If you need the whole workflow offline, test the speech model and the cleanup model independently.

Voconly publishes this comparison. It uses current project documentation and includes a repeatable offline trial for the mode you intend to keep.

Where each tool fits

Which settings should you check first?

Handy. Start with plain dictation. Try both the hold and toggle gesture in a local editor. If you use Linux, check the project's notes for your desktop environment, since global shortcuts and text insertion can require extra setup under Wayland. If Handy shows the correct transcript in its history but inserts different text, the project's troubleshooting notes explain how to check its clipboard paste method. Handy setup and troubleshooting

OpenWhispr. In AI Models → Speech-to-Text, select the Dictation tab, choose Local, and download a speech model. Next, check AI Models → Language Models → Dictation Cleanup. Turn cleanup off for a plain transcript or choose a local cleanup model for offline editing. A successful offline paste can be the fallback when cleanup fails, so check whether the cleanup itself actually ran. OpenWhispr lists unlimited local dictation on its free tier and a separate quota for its hosted transcription. OpenWhispr local models · Cleanup behavior · Pricing

Voconly. Start with a local speech model and check the plain transcript. For a second workflow, assign a scenario to another shortcut and connect it to a local Ollama model if you want the cleanup to run offline. If you connect an online language model instead, that provider can receive the transcript for processing. Voconly project documentation · Voconly features

Can you still get the output you need with Wi-Fi off?

1. Download the models first. In each app, select your local speech model and turn off cleanup for the first pass. If the same model is available across all candidates, start there. Otherwise, record the model name next to each result.
2. Repeat one phrase in a local editor. With the same microphone and the network off, say “Leave batch B seventeen pending until I hear back from Inez.” Check the name, batch number, and condition. A paste in the wrong place is an insertion problem. Missing or incorrect words are a recognition problem.
3. Test editing independently. For the tools you would use with cleanup, run the same phrase with “um” at the start. Check that the filler disappears, the condition survives, and the app confirms the cleanup step ran. Record the output from both passes.

Choose based on the stage you actually need to improve. If the correct transcript rarely needs editing, compare the shortcut and text insertion. If you frequently speak rough drafts, compare how much editing remains after the cleanup step and whether that step runs offline.