Support
Need help? Email us at support@thoth-app.com
Getting Started
What can I do with the free version?
The free version lets you record from your microphone or import audio files. You get up to 5 recordings, each up to 15 minutes long. Transcripts can be exported as plain text.
What does Thoth Pro unlock?
Pro removes all limits and unlocks the full feature set:
- Unlimited recordings with no time cap
- System audio capture and mixed recording (mic + system audio simultaneously)
- Speaker diarization — color-coded transcripts that identify who said what
- AI enhancements — improve, summarize, and translate your transcripts
- Large V3 Turbo transcription model for near-perfect accuracy
- All export formats: PDF, RTF, Markdown, JSON, M4A, AAC
What are the system requirements?
Thoth requires macOS Tahoe (26.0) or later. Since macOS 26 only runs on Apple Silicon, any Mac that can run it is supported. The Large V3 Turbo transcription model requires M2 or later; the compressed variant works on M1 or later.
Recording
How do I capture both my microphone and system audio at the same time?
Use Mixed recording mode (Pro). Thoth handles both channels natively — no third-party virtual audio drivers required.
I get a warning about a 4 GB file size limit. What does that mean?
WAV is an uncompressed format. Long recordings at high quality can approach the 4 GB limit on some filesystems. Thoth shows a warning after 2 hours 30 minutes. If you need to record longer sessions, split them into multiple shorter recordings.
Can I import existing audio files?
Yes. Use Import mode to select an audio file and run transcription on it.
Transcription
Which transcription model should I use?
It depends on your priority:
- Base — fastest, good for quick notes and drafts
- Small — balanced speed and accuracy, great for everyday use
- Medium — high accuracy, recommended for interviews and meetings (Pro)
- Large V3 Turbo — near-perfect accuracy at fast speeds, requires M2 or later (Pro)
- Large V3 Turbo Compressed — 4-bit quantized variant, works on M1, slightly lower accuracy (Pro)
The transcription model fails to load when I re-transcribe a recording
This is a known issue with the Large V3 Turbo model. If the model fails to load during re-transcription, quit Thoth completely and reopen it. The model will load correctly on the next launch. We are working on a fix.
How do I improve transcription accuracy?
A few things help significantly:
- Set the language manually if you know it, rather than relying on auto-detection
- Use a higher-quality model (Medium or Large V3 Turbo)
- Record in a quiet environment with minimal background noise
How many languages are supported?
Thoth supports 99+ languages for transcription, including English, French, Spanish, German, Chinese (Mandarin), Japanese, Korean, Portuguese, Russian, and many more. Language can be set per recording or detected automatically.
Where are the transcription models stored?
Models are downloaded from HuggingFace on first use and cached in your Documents folder. They are never sent anywhere — all inference runs locally on your Mac.
AI Features
What AI enhancements are available?
Pro users can apply four types of AI enhancement to any transcript:
- Improve — corrects errors, fixes punctuation, and improves readability
- Improve Diarization — corrects speaker attribution errors based on conversational context
- Summarize — generates a structured summary with overview, key points, decisions, action items, and notable quotes
- Translate — translates the transcript to any target language
Do AI features require an internet connection?
Transcription is always fully offline. For AI enhancements (improve, translate, improve diarization), a cloud API key is required — these always need an internet connection. Summarization is the exception: it can run fully on-device using a local model, or via a cloud API key if you prefer.
Which local AI models are available for summaries?
Local summarization runs on-device using Llama or Qwen models via llama.cpp — no server, no account needed:
- Fast model — Llama 3.2 3B (~2.3 GB): quick, lighter on resources
- Quality model — Qwen 2.5 7B (~5.1 GB): recommended for best results
Models are downloaded once from HuggingFace and run entirely on your Apple Silicon GPU.
How do I use cloud AI (OpenAI, Anthropic, Google)?
Go to Settings → AI and enter your API key. Your key is stored securely in the macOS Keychain. Thoth sends only the transcript text to your chosen provider — no audio is ever transmitted.
Export
What export formats are available?
- Free: plain text (.txt), WAV audio
- Pro: plain text, Markdown, RTF, JSON (with timestamps and speaker labels), PDF (formatted with speaker identification), M4A, AAC
Subscription & Billing
What plans are available?
- Monthly — $9.99/month, includes a 1-month free trial
- Annual — $79.99/year, includes a 1-month free trial (saves ~33%)
- Lifetime — $149.99 one-time purchase
How do I cancel my subscription?
Subscriptions are managed entirely by Apple. Open System Settings → Apple ID → Subscriptions, find Thoth, and cancel from there.
How do I restore my purchase on a new Mac?
Open Thoth and go to Settings → Subscription → Restore Purchases. Make sure you are signed in to the same Apple ID used for the original purchase.
Is there a family sharing option?
Not at this time. Each subscription covers a single Apple ID.
Privacy
Does Thoth send my recordings to the cloud?
No. All transcription runs on your Mac using on-device AI models. Your recordings and transcripts never leave your device unless you explicitly use a cloud AI feature with your own API key — in which case only the transcript text is sent, never the audio.
Does Thoth collect any analytics or usage data?
No. Thoth collects nothing. There are no analytics, no crash reporters, no telemetry. We don’t run any backend server — the only server we operate is this website. Everything else happens on your Mac. See our Privacy Policy for full details.
Contact
If your question isn’t answered here, or if you’ve found a bug, reach out at:
We typically respond within 1–2 business days.