How It Works
The path your voice takes through SpeakUp is simple and entirely local:
- You press your hotkey. SpeakUp begins capturing audio from your microphone.
- You speak. Audio is buffered in memory on your Mac.
- You press your hotkey again. Recording stops.
- whisper.cpp processes the audio. The Whisper model runs on your Mac's GPU via Metal acceleration. No network call is made.
- Text appears at your cursor. SpeakUp simulates keyboard input to type the transcription into whatever app is focused.
- Audio is discarded. The audio buffer is released from memory. No recording is saved anywhere.
At no point does any data leave your Mac. There is no server to send it to, no API to call, no cloud infrastructure of any kind.
What We Do Not Collect
- No audio recordings
- No transcriptions
- No usage data or analytics
- No crash reports
- No telemetry
- No account information
- No cookies (in the app)
- No email addresses
- No device identifiers
This is not a list of things we promise not to share. It is a list of things that do not exist. We cannot access what we never collect.
What We Do Not Know About You
Everything. We have no user database. We have no email list from app users. We have no usage metrics from the app. We do not know how many times you dictate per day, what language you use, how long your sessions are, or what apps you dictate into.
When you buy SpeakUp, Creem (our payment processor) handles the transaction. They know your payment details. We receive a license key and a confirmation. We do not receive your name, email, or any personal information from the app itself.
Cloud Dictation: The Privacy Risk
Cloud-based dictation tools follow a different architecture:
- Your voice is recorded by the microphone.
- The audio is compressed and sent over the internet to a remote server.
- The server runs speech recognition and returns the text.
- The audio may be stored for quality improvement, model training, or legal compliance.
At every step after step 1, your voice data exists on infrastructure you do not control. Server employees may have access. Security breaches can expose it. Government requests can compel its release. Terms of service may allow its use for AI training.
This is not a theoretical risk. Major cloud providers have confirmed that human reviewers listen to voice assistant recordings. The only way to eliminate this risk is to eliminate the cloud entirely.
Wispr Flow's "Privacy Mode"
Wispr Flow offers a Privacy Mode toggle in their settings. When enabled, they say your audio is not stored after processing. When disabled, your audio may be used to improve their service.
SpeakUp does not have a Privacy Mode because it does not have a non-private mode. There is no toggle because there is no cloud to toggle. The architectural difference is not a matter of degree -- it is a matter of kind. They offer a promise. We offer an impossibility.
GDPR/DSGVO Compliance
The General Data Protection Regulation requires organisations to protect personal data they collect and process. SpeakUp complies with GDPR in the most thorough way possible: by not collecting or processing any personal data at all.
There is no Data Processing Agreement (DPA) to sign because there is no data processing. There is no data deletion request to honour because there is no data to delete. There is no data breach to report because there is no data to breach.
For organisations in regulated industries -- healthcare, legal, finance, government -- this architecture eliminates an entire category of compliance burden. Your data protection officer does not need to evaluate SpeakUp's data handling practices because there are no data handling practices to evaluate.
For our full legal privacy policy, see Privacy Policy.
Related: SpeakUp Features · SpeakUp vs Wispr Flow · Dictation for Doctors · Dictation for Lawyers