We build intelligent, privacy-first software for professionals who demand precision and control. Every product runs on-device — your data stays yours.
Private, on-device transcription for macOS. Import any recording, label speakers, and generate intelligent summaries — all processed locally on your Mac.
Download on the Mac App Store
macOS 14.6+ · Apple Silicon (M1 or later)
Import audio or video. AI-powered speech recognition transcribes with high accuracy across 60+ languages. AI summaries available in English and major languages.
Automatically detect and label speakers. Rename a speaker once — all phrases in that file update. With Pro+, the same voice is recognised across multiple recordings.
Generate structured summaries tailored to meetings, interviews, podcasts, and more — with key decisions, action items, and speaker contributions. On-device AI — no cloud, no data sharing.
Built-in speech enhancement, background noise removal, and silence removal. Better input, better output.
Export as TXT, Markdown, HTML, or SRT subtitles — all with speaker names. Ready for your workflow.
Zero data collection. Zero analytics. Zero cloud processing. Your recordings and transcripts never leave your Mac.
| Vox Dictum | MacWhisper | Otter.ai | Dragon | |
|---|---|---|---|---|
| On-device processing | ✓ | ✓ | ✗ | ✓ |
| Zero data collection | ✓ | ✓ | ✗ | ✗ |
| AI Summary | ✓ On-device | ✗ | Cloud | ✗ |
| Speaker recognition | ✓ On-device | ✗ | Cloud | ✓ |
| Overlap detection | ✓ | ✗ | ✗ | ✗ |
| Audio enhancement | ✓ | ✗ | ✗ | ✗ |
| Free tier | Unlimited | ✗ | 300 min/mo | ✗ |
| Price | From £7.99/mo | €59 one-time | $16.99/mo | $699 |
Popular
Pro
£7.99 /month
Everything in Free, plus:
Advanced transcription models
AI Powered Summary
Overlap resolution
Custom vocabulary corrections
Speaker reallocation
Transcript phrase splitting
Bulk transcript export
🔒 Cobalt InFX products collect no data. Recordings, transcripts, and summaries are processed entirely on your device and are never transmitted to any server.
Vox Dictum processes audio and video recordings to produce text transcripts, speaker labels, and AI-generated summaries. All processing is performed locally on your Mac using on-device machine learning models. No audio, text, or metadata is sent to Cobalt InFX, Apple, or any third party during processing.
Nothing. Vox Dictum does not collect personal data, usage analytics, crash reports, telemetry, or any other information from your device. The app contains no analytics frameworks, no tracking pixels, and no third-party SDKs that collect data.
The only network activity in Vox Dictum is: (1) down AI models on first use (~2 GB, one-time), (2) down additional transcription models when you select a new model size, and (3) Apple verifying your subscription status. No recording data, transcript content, or user-generated content is transmitted over the network at any time.
Speaker voice matching (Pro+ tier) uses on-device neural network inference to compare voice characteristics within a single processing session. Voice embeddings are computed transiently in memory and discarded when the processing job completes. No biometric data is stored persistently. No voice profiles are created or retained between sessions.
Subscriptions are managed entirely by Apple through the App Store. Cobalt InFX does not process payments, store credit card information, or have access to your Apple ID credentials.
Your transcripts, speaker names, vocabulary corrections, and summaries are stored locally in your Mac's Application Support directory within the app's sandbox. This data is included in Time Machine backups and is deleted when you uninstall the app. Cobalt InFX has no access to this data.
Cobalt InFX products are not directed at children under 13. We do not knowingly collect any information from children.
If we update this policy, we will post the revised version on this page with an updated effective date.
Effective date: April 2026
Contact: support@cobaltinfx.com
Check the built-in Help section in Vox Dictum: Settings → Help. Most common questions are answered there.
Model downloads require a stable internet connection (~2–7 GB depending on features used). Transcription with advanced models may take longer than the recording duration — this is normal for on-device processing.
Click the button below to open your email client. Describe your issue and attach screenshots if relevant.
Please include:
support@cobaltinfx.com