Thoth

Product Introduction

1. Definition: Thoth is a 100% native macOS application designed for high-performance, private audio transcription and AI-driven summarization. It functions as an on-device AI scribe, utilizing local hardware acceleration via Apple Silicon to process audio without requiring cloud-based servers. Technically, it integrates OpenAI’s Whisper model through WhisperKit and CoreML to provide industrial-grade speech-to-text capabilities directly on the user's local machine.

2. Core Value Proposition: Built by a Laser Physicist to reclaim the dormant computational power of modern Macs, Thoth exists to eliminate the privacy risks and recurring costs associated with cloud-based transcription services. By processing data locally, it ensures that confidential conversations, intellectual property, and sensitive meetings never leave the user's hardware. It targets the "Privacy-First" market, offering a robust alternative to SaaS models that harvest user data for training or surveillance.

Main Features

1. Mixed Audio Routing & System Capture: Unlike traditional recording tools that require clunky third-party virtual drivers (like BlackHole or Soundflower), Thoth features a built-in Core Audio routing engine. This allows users to capture both microphone input and system audio (e.g., participants in a Zoom, Microsoft Teams, or Google Meet call) simultaneously. This is achieved through native macOS APIs, ensuring low latency and high-fidelity captures without complex configuration.

2. On-Device Whisper & CoreML Transcription: Thoth utilizes WhisperKit to run transcription models locally on the Mac’s Neural Engine. By optimizing these models for CoreML, the app achieves rapid processing speeds that rival cloud APIs. Users can choose between different model sizes depending on their hardware and accuracy requirements, ensuring high-precision speaker identification and timestamping without an internet connection.

3. Local LLM Summarization Engine: For post-transcription analysis, Thoth offers an on-device AI summarization suite. Users can download and run local Large Language Models (ranging from 1.9 GB to 6.8 GB) to generate insights, action items, and summaries. This local inference ensures that the context of a transcript is never exposed to third-party AI providers.

4. SwiftUI & AppKit Native Architecture: Thoth is built entirely in SwiftUI and AppKit, eschewing cross-platform frameworks like Electron. This results in a significantly lower memory footprint, support for native macOS features like Dark Mode and collapsible sidebars, and fluid animations. This native approach ensures that the app remains performant even when handling large batch imports of audio files.

5. Keychain-Secured "Bring Your Own Key" (BYOK): While Thoth prioritizes local processing, it offers flexibility for users who prefer cloud LLMs like GPT-4, Claude, or Gemini. The app implements a secure BYOK architecture where API credentials are stored in the Apple Keychain—the most secure local encrypted storage on macOS. Requests are sent directly from the app to the provider, with no intermediary servers.

Problems Solved

1. Privacy & Data Sovereignty: Many professionals in legal, medical, or corporate sectors are prohibited from using cloud transcribers due to compliance regulations (GDPR, HIPAA, etc.). Thoth solves this by keeping 100% of the data on the local disk, removing the "data harvesting" risk inherent in cloud-based "free" or "pro" transcription services.

2. The "Rental" Economy & Server Costs: Traditional transcription services charge per minute or through high monthly subscriptions to cover their server costs. Thoth shifts the cost to the user's existing hardware, allowing for a "Buy Once, Use Forever" model (Lifetime Pro) that eliminates the financial burden of renting cloud GPUs.

3. Complex Audio Setups: Capturing meeting audio on a Mac usually involves complex routing software that often breaks during OS updates. Thoth simplifies this with its native integration, making it a "one-click" solution for recording both sides of a digital conversation.

4. Target Audience:

• Legal & Medical Professionals: Who require absolute confidentiality for client/patient recordings.
• Journalists & Researchers: Conducting sensitive interviews that cannot be uploaded to third-party servers.
• C-Suite Executives: Recording board meetings and internal strategy sessions.
• Developers & Power Users: Who prefer native macOS performance and the ability to leverage their Mac’s Neural Engine.

5. Use Cases:

• Confidential Meeting Minutes: Recording and summarizing high-stakes Zoom or Teams calls locally.
• Academic Lectures: Batch importing voice notes and lectures for automated, color-coded speaker transcription.
• Offline Content Creation: Transcribing podcasts or video scripts in environments without internet access.
• Archive Management: Exporting meeting transcripts into searchable formats like Markdown or JSON for personal knowledge management systems (PKMs).

Unique Advantages

1. Differentiation from Cloud Competitors: Most competitors (e.g., Otter.ai, Fireflies) act as "bots" that join meetings and upload audio to the cloud. Thoth is a "silent" local tool. It does not require an account, does not join meetings as an external participant, and does not require a persistent internet connection. It is an "invisible" and private workflow enhancement.

2. Hardware Optimization: By requiring macOS Tahoe and recommending Apple Silicon, Thoth is specifically tuned for the latest M-series chips (M1, M2, M3, M4). This ensures that the local AI models run with maximum efficiency, utilizing the unified memory architecture of Apple Silicon to handle large audio files that would crash standard Electron-based apps.

3. Comprehensive Export Options: Thoth goes beyond simple text files, offering professional-grade exports including PDF, RTF (for Word compatibility), Markdown (for Obsidian/Logseq users), and JSON (for developers). All exports include precise timestamps and speaker labels.

Frequently Asked Questions (FAQ)

1. Is Thoth really 100% private and offline? Yes. Thoth is designed with a "Zero Upload" philosophy. The core transcription engine (WhisperKit) and the local AI summary models run entirely on your Mac’s CPU, GPU, and Neural Engine. No audio data or transcript text is ever sent to Thoth's servers.

2. Does Thoth require a subscription to work? Thoth offers a "Free-to-try" model with a generous Pro tier. While there are monthly and yearly options, it also offers a "Lifetime" license, allowing users to pay once and use their own hardware for transcription indefinitely, avoiding the "SaaS trap."

3. Can Thoth record audio from Zoom or Microsoft Teams? Absolutely. Thoth features native mixed audio routing that captures both your microphone and the audio output from applications like Zoom, Teams, and FaceTime without needing external virtual audio cables or drivers.

4. What are the system requirements for Thoth? Thoth requires macOS Tahoe or later. While it runs on Intel-based Macs, Apple Silicon (M1/M2/M3/M4 chips) is highly recommended for the best performance, as the AI models are optimized for the Apple Neural Engine.

5. Can I use my own AI models or API keys? Yes. Thoth allows you to run local models (up to 6.8 GB) for summaries. If you prefer high-end cloud models, you can provide your own API keys for OpenAI, Anthropic, or Google. These keys are stored securely in your macOS Keychain and never shared with the developer.