Offline dictation on a Mac
Offline dictation means the speech model sits on your own disk and your Mac does the recognition itself. On Apple Silicon this is practical; on Intel Macs it mostly is not. This guide explains what happens under the hood, what to check before you rely on it on a plane or with confidential notes, and how to prove to yourself, with Wi-Fi switched off and with an outbound firewall, that audio is not leaving the machine. We make Voice Monkeys, which works this way on Apple Silicon, but every check below works with any app.
What offline actually means
A speech recognition model is a large file of learned weights. In a cloud setup your microphone audio travels to a server, the model runs there, and text comes back. In an offline setup the same kind of model is stored on your Mac, the audio goes from the microphone straight into it, and the text is produced locally. Nothing needs to travel for the words to appear.
Apple Silicon chips (M1 and newer) include a Neural Engine built for exactly this kind of maths, which is why apps that transcribe locally tend to require them. The same model can run on an older processor, but slowly enough that vendors often steer Intel users to the cloud instead.
Offline describes where recognition happens. It does not automatically mean an app sends nothing at all: it could still check for updates, verify a licence or send usage data. The two tests further down separate those questions.
Which Mac dictation tools run on the device
| Tool | On the Mac? | What the vendor states |
|---|---|---|
| macOS Dictation | Depends | Keyboard settings show whether dictation is processed on the device or sent to Siri servers |
| Voice Monkeys (Apple Silicon) | Yes | Speech model on the Neural Engine, offline after a one-time ~600 MB download |
| VoiceInk | Yes | Transcription on device with local models; optional cloud enhancement handles only the transcribed text |
| Superwhisper | Offline models available | Offline models only run really well on Apple Silicon Macs |
| Wispr Flow | Not stated | The pricing page does not say where speech is processed |
For Apple's built-in, open System Settings > Keyboard > Dictation and read the line under the switch; it answers the question for your Mac and language. Setup details are in the Mac dictation guide, and Voice Monkeys vs macOS Dictation compares the two directly.
The one-time model download
Every offline app needs the internet once, to fetch its model. For Voice Monkeys that is about 600 MB; other apps offer models of different sizes, and bigger is not automatically better for your voice. Until the download has finished, the app cannot work offline, so do this at home, not at the gate.
- Check free space first: System Settings > General > Storage. Leave a few gigabytes spare beyond the model size.
- Use a stable connection and let the download finish; do not close the lid halfway.
- Dictate one sentence while still online to confirm the model loads.
- Then run the Wi-Fi test below once before trusting it on a flight or in a place without signal.
Voice Monkeys on Apple Silicon (M1 or newer, macOS 14+) costs $25 once and has a 7-day trial without a card; download it and the model is fetched on first start. If you want other local options, see VoiceInk alternatives and the best dictation apps for Mac.
The Intel Mac caveat
Intel Macs have no Neural Engine. Superwhisper's site says Intel Macs work best with its cloud models. Voice Monkeys on Intel runs as a cloud plan ($12 a month or $96 a year): audio is sent encrypted to a server in Germany (Hetzner, Falkenstein), turned into text in memory and discarded, with neither audio nor text stored, only the seconds counted. That is a clear, documented path, but it is not offline, and nobody should describe it as such.
To see your chip, open Apple menu > About This Mac. A Chip line starting with Apple means Apple Silicon; a Processor line naming Intel means an Intel Mac. On Intel, the built-in Dictation is still worth checking: read its Keyboard settings line for your language.
Test 1: switch the network off
- Open the dictation app and confirm its model has finished downloading.
- Click Wi-Fi in Control Center and turn it off. Unplug any Ethernet cable and disconnect any phone hotspot.
- Open TextEdit and dictate: “Flight leaves at 6:40 from gate B12, call Tomas when you land.”
- If the text appears, recognition happened on your Mac. If you get an error or nothing, the app depends on a server.
- Dictate for a couple of minutes more, then switch Wi-Fi back on.
This proves the words were produced locally. It does not prove the app never sends anything later, for example queued audio or text once the connection returns. That is what the second test is for.
Test 2: watch outgoing connections
An outbound firewall such as Little Snitch shows every connection an app tries to make, with the destination, and lets you allow or block it. Run it while you dictate. A local app should not open a connection when you press the dictation key; if a connection appears exactly when you speak and carries a steady stream of data, audio or text is going somewhere.
Without extra software, open Activity Monitor, choose the Network tab, sort by Sent Bytes and find the dictation app. Note the number, dictate for one minute, and look again. A minute of audio is a sizeable upload, so a local app's counter should barely move. In Terminal, `nettop` gives the same per-process view live.
- Small connections at launch are often update or licence checks. If you are unsure, ask the vendor what each destination is.
- Block the app entirely in the firewall, then dictate. If it still works, recognition does not need the network.
- For Apple's Dictation, also check the optional setting about sharing audio recordings with Apple and set it the way you want.
None of this replaces your organisation's own rules. If you dictate patient, client or legal material, these tests tell you where speech is processed, and your employer or regulator decides whether that is acceptable.
A short checklist before relying on it
- Apple Silicon confirmed in About This Mac.
- Model downloaded and tested once with Wi-Fi off.
- Outgoing traffic checked in a firewall or Activity Monitor while dictating.
- Audio sharing options in the app reviewed.
- Your rules for sensitive material checked with whoever sets them.
If cost is part of the decision, dictation apps without a subscription lists which local apps you can buy once, and Voice Monkeys vs Superwhisper compares a pay-once app with one that has a free tier.
Questions and answers
Does Mac dictation work without internet?
It can. Apple's Keyboard settings show whether Dictation runs on the device for your Mac and language. Apps such as Voice Monkeys and VoiceInk transcribe on Apple Silicon Macs after a one-time model download.
How big is an offline dictation model?
It varies by app and model. The Voice Monkeys model for Apple Silicon is about 600 MB, downloaded once. Check free space under System Settings > General > Storage first.
Can I dictate offline on an Intel Mac?
Usually not well. Superwhisper says its offline models only run really well on Apple Silicon, and Voice Monkeys on Intel uses a cloud plan. Check what Apple's built-in Dictation settings say for your language.
How do I know my dictation app is not sending audio?
Turn off Wi-Fi and dictate: if text appears, recognition is local. Then watch the app's outgoing traffic with an outbound firewall such as Little Snitch, or with Activity Monitor's Network tab, while you speak.
Is offline dictation automatically compliant for medical or legal work?
No tool's offline mode is a compliance guarantee. It tells you where speech is processed; whether that meets your obligations is for your employer or jurisdiction to decide.
Sources
- Apple: Use Dictation on Mac (checked November 6, 2026)
- VoiceInk (checked November 6, 2026)
- Superwhisper (checked November 6, 2026)
- Wispr Flow pricing (checked November 6, 2026)