That is not automatically a flaw. Cloud dictation can be fast, accurate, and easy to set up. But it does mean Wispr Flow is a poor fit for people who need dictation to work without a network connection, behind restrictive firewalls, on flights, or in workflows where voice transcription should happen on the Mac itself.

If offline or local transcription is the point, you want a different architecture.

What Wispr's docs say

Wispr Flow's help center describes internet connectivity as a requirement for voice transcription. Its troubleshooting pages also cover cases where VPNs, proxies, security tools, offline state, or network failures prevent dictation from working normally.

That tells you something important: Wispr Flow is designed around a cloud transcription path. Privacy and retention settings can affect what is stored after processing, but they do not turn the product into a local transcription engine.

Offline mode vs privacy mode

These two ideas are easy to mix up:

  • Offline transcription means audio is processed on your device without sending it to a remote transcription service.
  • Privacy or zero-retention mode means the service processes data remotely but limits what is stored after processing.

Both can be useful. They solve different problems.

If your requirement is "do not keep my transcript after processing," a privacy mode may be enough. If your requirement is "do not send this audio off my Mac in the first place," you need local transcription.

Why local transcription matters

Local transcription is useful when:

  • You dictate in places with unreliable internet.
  • You work behind VPNs, proxies, or strict security tools.
  • You handle sensitive drafts, client notes, or internal work.
  • You want your dictation workflow to keep working during cloud incidents.
  • You want to choose the model and endpoint yourself.

It is not magic. Local models can be slower or less polished depending on your hardware and model choice. But the tradeoff is control.

How Foil handles local dictation

Foil supports local whisper.cpp through a localhost OpenAI-compatible server. In practice, that means Foil can send audio to a transcription service running on your own Mac instead of requiring a hosted dictation provider.

Foil also supports hosted providers, including Groq Whisper and OpenAI Whisper, plus custom OpenAI-compatible endpoints. The point is not that local should be the default for everyone. The point is that the route is your choice.

A practical decision rule

  • Choose cloud dictation when you value polish, convenience, and cross-device availability more than local control.
  • Choose local dictation when availability, privacy architecture, or endpoint control matters more than a managed all-in-one writing layer.
  • Choose a provider-flexible tool when you want to move between both modes.

Foil is in the third category. You can start local, use hosted transcription when it makes sense, and keep custom endpoints available for workflows that already have trusted infrastructure.

Try a local-first Mac dictation workflow

If your search started with "does Wispr Flow work offline?", the useful next step is not another generic comparison. It is a quick local transcription test.

Install Foil, configure local whisper.cpp, dictate into the app you already use, and see whether the local route is good enough for your everyday work.

Sources and further reading