Transcribe · Summarize · Zero Upload
Your Private AI Scribe
Record both sides of any meeting. Transcribe and summarize locally on your Mac. No cloud, no data ever leaving your machine.
Requires macOS Tahoe · Apple Silicon recommended · Free to try
The Scribe's Craft
Capture Both Sides of the Conversation
Record your microphone and system audio at the same time. No third-party routing tools, no virtual audio drivers. Just hit record.
Built for Mac
Native SwiftUI
Real macOS app. Collapsible sidebars, smooth animations, proper Dark Mode. Runs on AppKit, not Electron.
Export Anywhere
PDF, Markdown, RTF, or JSON with timestamps and color-coded speakers. Share or archive however you want.
99 Languages
Auto-detects the language and transcribes it. English, French, Japanese, Arabic โ Whisper handles them all, entirely on your Mac.
Audio Architecture
Two channels.
Zero ambiguity.
Most tools blend your mic and remote audio into a single stream, then guess who spoke when. Thoth keeps them separate from the very first sample. No guesswork needed.
Channel 1 · Mic
Your Voice
Dedicated stream from your microphone. Captured independently and identified as you before transcription starts. Never mixed with remote audio.
Always Speaker 1Channel 2 · System
Remote Voices
Captured from Zoom, Teams, Meet or any app via the macOS audio engine. A completely independent stream. No screen capture, no bot joining the call.
Speaker 2 and beyondNo guesswork. Every word attributed correctly.
When channels are separate, speaker attribution for remote participants is deterministic, not estimated. Mic is always you. System audio is always them. Even when everyone talks at once.
Absolute Privacy
Under the Scribe's Seal.
Your data stays yours. Built from day one for people who handle confidential information.
100% Offline Engine
Transcription runs on your Mac through WhisperKit and CoreML. No audio ever leaves your machine.
Local Speaker Detection
An on-device engine detects who is speaking and color-codes your transcript. No cloud processing involved.
On-Device AI Summaries
Choose from five local models (1.9 GB to 6.8 GB) and run AI summaries entirely on your Mac. No API key, no cloud.
Bring Your Own Key
Prefer OpenAI, Anthropic, or Google? Use your own API keys. Requests go direct from your Mac to the provider.*
Keychain Secured
API credentials stay locked in your Apple Keychain. Thoth never stores keys in plaintext or sends them anywhere.
Under the Cartouche
Real numbers, real hardware.
Benchmarked on a 42-minute recording, Apple M2 MacBook Pro.
Transcription Performance
- 3.3 min to transcribe 42 min of audio 12.7ร realtime
- All processing on Apple Neural Engine via CoreML
- 99 languages with auto-detection
Large V3 Turbo model
Diarization Performance
- 7.72 seconds for 42 min of audio with 2 speakers
- Up to 8 speakers supported
- Mixed audio (Zoom, Teams, Meet): attribution is deterministic. Mic and system audio are separate streams, no ambiguity
Fully on-device via Pyannote CoreML
Local vs Cloud AI Summaries
Tested on a real French-language interview transcript. Scored by Claude Opus across 6 criteria.
Local ยท Qwen 7B
~5/10
Fast and private. Good for quick overviews. Struggles with nuanced decision capture and quote selection.
Cloud ยท Claude Sonnet (BYOK)
~8.7/10
Captures operational detail, adapts to content type, better quote selection.
| Local (Qwen 7B) | Cloud (Claude Sonnet) | |
|---|---|---|
| Factual accuracy | 7/10 | 9.5/10 |
| Completeness | 5/10 | 9/10 |
| Decision capture | 2/10 | 8.5/10 |
| Action items | 5/10 | 8/10 |
| Quote selection | 4/10 | 8.5/10 |
| Language quality | 7/10 | 9/10 |
| Overall | ~5/10 | ~8.7/10 |
| Privacy | Zero data leaves | Text sent to provider |
| Cost | Free | ~$0.01/hour |
| Internet required | No | Yes |
Honest takeaway: local models are best when privacy is non-negotiable or internet is unavailable. Cloud AI is better when depth matters.
The privacy guarantee stays constant regardless of which you choose. Audio never leaves your machine. If you use BYOK cloud AI, only the transcript text goes to your chosen provider, directly with your key. Thoth never sees it.
How We Compare
Local-first. Always.
Cloud recorders are convenient. They're also always listening.
| Thoth | Otter | Fireflies | Granola | |
|---|---|---|---|---|
| Audio stays on your Mac | โ | โ | โ | โ |
| No bot joins your call | โ | โ | โ | โ |
| Works fully offline | โ | โ | โ | โ |
| Dual-channel recording | โ | โ | โ | โ |
| On-device AI summaries | โ | โ | โ | โ |
| Native Mac app | โ | โ | โ | โ |
Competitor features are approximate and subject to change. Otter, Fireflies, and Granola are trademarks of their respective owners.
Pricing
Try free, then go Pro.
Start free with full transcription features. Upgrade for unlimited recordings and AI.
Free
$0
- 5 recordings
- 30 min (mic) / 15 min (system audio)
- 3 AI enhancements/month (local or cloud)
- WAV audio export
- TXT transcript export
Pro
- Unlimited recordings & duration
- System Audio & Mixed recording
- M4A, AAC, Markdown, RTF, JSON, PDF export
- Unlimited AI enhancements (local or cloud with your key)*
- Large transcription model
Free trial included with subscription
*Cloud AI features (OpenAI, Anthropic, Google) may not be available in all countries due to local regulations. On-device AI is available everywhere.
The Scribe Awaits
Private by default.
No cloud required.
Accurate transcripts and AI insights, running entirely on your hardware. No accounts, no uploads, no data leaving your Mac.