Transcribe · Summarize · Zero Upload
Press Kit
Thoth is a privacy-first macOS transcription app. Everything runs on your Mac. No cloud, no uploads, no accounts.
Last updated: May 2026
Download Press Kit
Quick Facts
About Thoth
Ready-to-use descriptions for editorial use. Copy and paste as needed.
Thoth is a macOS transcription app that records meetings, transcribes with WhisperKit or Parakeet, and generates AI summaries entirely on-device. No account, no cloud, no bot joining your calls. Works with Zoom, Teams, and any app. Turn off Wi-Fi: it still works. Built solo by a laser physicist in Bordeaux.
Thoth is a privacy-first meeting recorder for Mac that keeps the entire pipeline on your device. It captures system audio and microphone simultaneously using native Apple frameworks, no virtual driver required. Transcription runs locally via WhisperKit or Parakeet TDT, optimized for Apple Silicon. Speaker diarization identifies each voice on-device. AI summaries use five bundled local models, from Ministral 3B to Gemma 3 12B. The AI chat interface requires your own API key for OpenAI, Anthropic, or Google: requests go directly from your Mac to the provider, Thoth never sees them.
No account is required, no audio ever leaves the machine, and no telemetry is collected. The free tier includes unlimited recordings and 10 AI actions per month. Paid plans unlock longer sessions, larger models, and full AI access. Built solo by Matthieu Veinhard, a laser physicist based in Bordeaux, France.
Thoth is a macOS app that records, transcribes, and summarizes meetings entirely on your Mac. The privacy model is architectural: there is no server to breach, no company that sees your audio, and no cloud dependency to break. Turn off Wi-Fi and every feature works exactly the same.
Audio capture uses native Apple frameworks to record system audio and microphone simultaneously on separate channels. This means remote participants on Zoom, Teams, or Google Meet are captured cleanly through the system audio track, while your voice comes from the microphone, giving speaker diarization a deterministic foundation rather than an estimate. No browser extension, no virtual driver, no bot joining the call.
Transcription runs via WhisperKit or Parakeet TDT v3, both optimized for Apple Silicon Neural Engine. A floating Live Transcript panel shows text as it is spoken, with a fast preview model running in parallel with a quality refinement model. Re-transcription with a different model is one click.
AI summaries and action item extraction run on five bundled local models (Ministral 3B, Llama 3.2 3B, Gemma 3 4B, Qwen 2.5 7B, Gemma 3 12B). The AI chat interface uses your own API keys for OpenAI, Anthropic, or Google, with requests going directly from your Mac to the provider. Thoth never proxies these requests.
The free tier includes unlimited recordings and 10 AI actions per month. Pro plans unlock longer sessions, larger models, and unlimited AI actions. Export formats include PDF, Word, RTF, Markdown, plain text, and JSON, all generated locally.
Thoth launched on the Mac App Store in early 2026 and has been covered by MacGeneration. It is built and maintained solo by Matthieu Veinhard, a laser physicist (PhD) based in Bordeaux, France, with no external funding or team.
Selected Coverage
Record Any Audio
Capture system audio, microphone, or both at the same time using native Apple frameworks. Mic and system audio are kept as two independent channels, so speaker attribution for remote participants is deterministic, not estimated. Works with any app: Zoom, Google Meet, podcasts, lectures. No browser extension, no bot joining your call.
Live Transcript
A floating panel shows the transcript as you speak. Two models run in parallel: a fast one for instant preview and a quality one for the final text. Everything happens on-device with zero latency to the cloud.
Know Who Said What
On-device transcription via Whisper, optimized for Apple Silicon. Speaker detection identifies each voice and color-codes their contributions. No audio leaves your Mac.
Local AI, No Cloud Required
Five bundled AI models handle summaries, action items, and formatting right on your machine. Users who want cloud AI can bring their own API keys for OpenAI, Anthropic, or Google.
Instant Summaries
One click turns a raw transcript into a structured summary with key points, action items, and decisions. Export to PDF, Markdown, RTF, JSON, or audio formats.
AI Assistant
Ask questions about your recordings, extract specific details, or reformat transcripts. All processing happens locally unless you choose to use your own cloud API keys.
Most transcription apps send your audio to a server.
Thoth doesn't.
Turn off your Wi-Fi and it works exactly the same. Meeting recordings, medical appointments, legal consultations, private conversations: everything stays on your hardware. No account required, no telemetry, no data collection.
Bundled AI Models
Pricing
Free trial included. A free tier is also available with unlimited recordings and 10 AI enhancements per month.
Technical Requirements
Media Assets
All assets below are free to use in editorial coverage of Thoth.
About the Developer
Thoth is built by Matthieu Veinhard, a laser physicist (PhD) based in Bordeaux, France. His career has spanned attosecond physics research, coherent beam combining of 61-channel femtosecond lasers, and industrial R&D on high-power laser systems, with published work in Optics Letters and SPIE conferences along the way. Building complex instrumentation software in Python and LabVIEW has always been part of the job.
Thoth started as a personal tool. He kept running into the same problem: every meeting transcription app wanted his audio on a server. So in late 2025, he built his own: solo, no team, no investors. Every feature was added because he needed it himself. The privacy model isn't marketing; it's an architectural constraint baked in from the first commit.
Contact: [email protected] · LinkedIn
Need a promo code? Contact [email protected] for review copies.
