Cloud Mode vs On-Device Mode

Cloud Mode sends your audio to Groq through Walkie's Worker for transcription, with optional AI Polish. On-Device Mode runs the speech model on your computer and can polish locally with its optional intelligence pack: dictation audio never leaves, it works offline with no account, and it is unlimited and free forever. Both modes are available on every plan, including Free — this is a privacy choice, not a paid upgrade.

Side by side

Cloud ModeOn-Device Mode
Where audio is processedGroq, through Walkie's WorkerYour machine
Needs internet during speech processingYesNo, after model download
Removes filler wordsWith AI PolishWith optional local AI Polish
Adds punctuation and capitalisationYesYes; local AI Polish is optional
Formats for the app you are inWith AI PolishWith optional local AI Polish
Uses your custom dictionaryYesYes
Cost on Free2,000 words a weekUnlimited, forever
Cost on paid plansUnlimitedUnlimited

What each mode actually sends

Cloud Mode

Cloud Mode uploads the audio of a single dictation through Walkie's Worker to Groq and returns text. With AI Polish on, it can also send a destination-app category, selected text, up to 500 characters around the cursor on either side, or the sentence being rewritten. With AI Polish off, it sends audio without that text context. Walkie does not store the cloud clip after processing; Cloud Mode text history syncs to your account. Groq Zero Data Retention is enabled for inference, and providers are bound not to train on your content.

On-Device Mode

On-Device Mode runs speech recognition locally, so dictation audio and text do not leave your computer for processing. The app can still contact Walkie for model downloads and its signed continuity check; signed-in settings sync and opt-in analytics are separate. This is the mode to use for speech you are not permitted to send to a third party.

Choosing a mode

Use On-Device Mode when:

  • You are working with confidential, regulated, or client-privileged material
  • You are offline — on a plane, on a train, or behind a restrictive firewall
  • Your employer's policy forbids sending audio to third-party services

Use Cloud Mode when:

  • You want the fastest, cleanest result with no editing afterwards
  • You are dictating long-form prose where punctuation matters
  • You want text formatted differently in a terminal than in an email

Choosing an on-device model

On-Device Mode runs a speech model you download inside the app. Larger models are more accurate and slower; smaller models are faster and lighter on memory.

ModelLanguagesBest for
Parakeet V325 EuropeanThe default — fast and accurate
Whisper Small99Other languages on older or memory-constrained computers
Whisper Medium99More accuracy than Small, still modest
Whisper Turbo99The most accurate option, in any language, if you can wait a few seconds
MoonshineEnglish onlyUS English on modest hardware

Language detection

Cloud Mode detects the language you are speaking automatically and returns text in that language — speak Spanish, get Spanish. The cleanup step stays in your language rather than drifting into English.

Frequently asked questions about cloud Mode vs On-Device Mode

Is On-Device Mode a free tier and Cloud Mode a paid one?

No. Both modes are on every plan. The Free plan limits Cloud Mode to 2,000 words a week, but On-Device Mode dictation is unlimited on Free forever. Paid plans lift the Cloud Mode cap. The free-forever guarantee covers dictation — meetings and Read Aloud have their own limits on Free, whichever mode records them.

Does On-Device Mode still use my custom dictionary?

Yes. Dictionary corrections are applied locally after transcription, so they work in both modes.

What happens if Cloud Mode cannot reach the internet?

Walkie falls back to your downloaded on-device model rather than failing. If no model has been downloaded, the dictation is kept on disk and offered back to you for retry.

Which mode is more accurate?

For raw word accuracy they are close, and on English the local Parakeet model is excellent. Cloud Mode wins on the finished result because of the cleanup step — punctuation, casing, and filler removal.

View as markdown
Was this helpful?