Everything Thoth does.

Capture the room, the call, or both at once.

• Microphone, system audio, or both on separate channels
• No virtual audio driver, no screen capture, no bot joining the call
• Your mic is always Speaker 1; system audio carries the remote participants
• Both channels recorded, transcribed and separated independently, so attribution is deterministic
• Menu bar mode: start and stop without switching windows
• A nudge when Zoom, Teams or Slack launches
• An offer to record when a meeting starts in your calendar
• Import audio files you already have
• Folders three levels deep and colour-coded tags
• Automatic AAC compression so the library stays small

Four engines, ninety-nine languages, none of it in the cloud.

• A floating live panel shows the text word by word as you speak
• Two models run in parallel: a fast preview and a quality final pass
• WhisperKit across all 99 languages, with automatic detection
• Parakeet TDT v3 for batch speed, around 190x real time Pro
• Parakeet EOU streaming word by word, around 160 ms latency Pro
• Apple Speech, on device, live or batch
• Five Whisper sizes, from Base to Large V3 Turbo
• Feed it a custom vocabulary from a CSV
• Smart editing highlights the words the engine was unsure about
• Re-transcribe with a different model in one click
• Works entirely offline once the model is downloaded

Speaker separation that is measured, not guessed.

• On-device speaker identification, Pyannote CoreML Pro
• The transcript is colour-coded automatically
• Up to 8 speakers in one recording
• 7.72 seconds to separate a 42-minute, two-speaker recording
• Aligned word by word rather than estimated
• Rename speakers and they update across the transcript
• An AI pass can correct attribution from conversational context Pro
• Runs offline, no cloud involved

Summaries, chat and rewriting, on your hardware by default.

• One click turns a recording into notes, action items and key decisions
• Five bundled local models, from Ministral 3B to Gemma 3 12B
• Apple Intelligence on supported devices
• Bring your own API key if you want a larger cloud model
• Or point Thoth at your own compatible server
• Or use a verified hardware enclave, attestation checked before every session
• Save your own prompts and rerun them on any recording
• Ask questions about a recording in a chat panel
• Improve punctuation and readability, or translate the transcript
• Action items become real Reminders entries, only the ones you approve

The audio has no path off your machine.

• Audio never leaves your device, on any tier
• If you use a cloud model, only transcript text is sent, with your own key
• Anonymization runs on device before anything is sent
• Detects people, places, organizations, emails and phone numbers
• Card numbers and IDs validated by checksum, plus patterns you define
• Every detected item is shown as a token, and you decide what leaves
• The mapping is reversed locally, so the answer reads naturally
• No account, no user database, no analytics, no crash reporters
• Everything works with the network switched off

Get it out of Thoth and into whatever you actually work in.

Across your devices, through your account, not ours.

• iCloud sync carries recordings, transcripts, summaries and notes
• It runs through your own iCloud account, never a Thoth server
• Send to Device pushes a recording from Mac to iPhone directly
• Over AirDrop or your local Wi-Fi, with no cloud round trip
• One Universal purchase covers Mac, iPhone and iPad

The same scribe, in your pocket.

• The same on-device engine as the Mac app
• Record, transcribe, identify speakers and run the same AI actions
• Import audio from any app
• A Live Activity and Dynamic Island keep the recording visible
• Stop or pause without opening the app
• On iPad the library and transcript sit side by side
• Microphone only: iOS gives no app access to another app's audio
• The interface is fully localised in English and French