On-device by design
Every transcription and cleanup step runs on the user's CPU/GPU. There is no inference server to breach.
Zero audio retention
Audio buffers are held only in RAM, only during the press-to-talk window, and zeroed on release.
No screen access
SpeakrStream does not request, capture, or OCR your screen. Confidential documents stay confidential.
Encrypted at rest
Local vocabulary, settings, and licence data are encrypted with the OS keychain on every platform.
UK / EU residency
Admin console hosted in the UK with optional Frankfurt failover. On-prem deployment available.
Audit-ready logs
Per-user, per-event log with CSV export. No transcript contents — just metadata.
Certifications
SOC 2 Type II
ISO 27001
Cyber Essentials Plus
GDPR / UK GDPR
Verify it yourself
In 5 minutes, your IT team can prove SpeakrStream transmits nothing.
- Install SpeakrStream on a test machine.
- Open Wireshark (or Little Snitch / GlassWire) and filter by the SpeakrStream process.
- Disconnect from the network. Confirm dictation still works.
- Reconnect. Dictate for 10 minutes. Observe zero outbound packets.