WhaleRead

Product Introduction

1. Definition: WhaleRead is a local-first, privacy-focused macOS application for bilingual reading and AI-assisted translation. It is a technical tool in the categories of offline AI translation software, digital EPUB reader, and private document processing.
2. Core Value Proposition: It exists to enable cross-language reading of personal eBook libraries (TXT, Markdown, EPUB) while ensuring data privacy and user control. Its core value is providing on-device AI translation and self-hosted large language model (LLM) integration to keep a user's entire library and reading data on their local machine or private infrastructure.

Main Features

1. Bilingual Reading Workspace: The app presents source text and AI-generated translation in a synchronized, side-by-side or switchable view. It preserves original EPUB formatting, chapters, and images. This workspace maintains a unified system for bookmarks, notes, and search across both language versions, creating a calm reading environment without switching apps.
2. Flexible, Private AI Model Deployment: Translation and "Ask AI" functions can run on two private paths. The on-device 7B parameter model performs inference locally on the user's Mac, ensuring zero data transmission. Alternatively, users can deploy a more powerful self-hosted 30B parameter model on their own hardware or private server (e.g., via Ollama, LM Studio) for higher quality, retaining full data sovereignty.
3. Evidence-First Review & Human-in-the-Loop Editing: Instead of automatic overwrites, all AI suggestions are treated as drafts. The review interface locks the source text, the current translation, the AI's reasoning for its suggestion, and the user's final edit together. This human confirmation workflow ensures no change is applied to the book file without explicit user approval, treating model fallibility as a core design parameter.

Problems Solved

1. Pain Point: It solves the privacy risk of cloud-based translation services (like Google Translate, DeepL) which require uploading entire book files. It also addresses the inconvenience and context loss of using separate translation tools and readers, and the poor idiom handling of traditional dictionary-based translation software.
2. Target Audience: Primary personas include privacy-conscious polyglot readers, language learners analyzing texts, researchers working with foreign-language documents, and professionals (e.g., lawyers, academics) who need to review materials in other languages but cannot risk exposing sensitive content to third-party APIs.
3. Use Cases: Essential for: Reading purchased EPUBs in a foreign language with immediate, private translation. Studying a novel in its original language with a parallel translation for comprehension. Preparing a local document collection for analysis without compromising confidentiality. Reviewing and refining AI-generated translations for nuanced, publishable quality.

Unique Advantages

1. Differentiation: Unlike consumer cloud readers (Amazon Kindle, Google Play Books) or translation APIs, WhaleRead operates on a local-first architecture. Unlike Calibre with translation plugins, it offers a deeply integrated, privacy-by-design workflow with a dedicated review system and support for running advanced private LLMs (30B parameters).
2. Key Innovation: Its dual-model private inference system bridges the gap between convenience and quality. The on-device 7B model offers a baseline of usable, entirely offline translation, while the self-hosted 30B model option provides near-cloud-quality output without data leakage. The immutable review ledger that ties source evidence to AI reasoning and human edits is a novel approach to human-AI collaboration for translation.

Frequently Asked Questions (FAQ)

1. What file formats does WhaleRead support for translation? WhaleRead supports local TXT files, Markdown (.md) files, and EPUB eBook files for AI-powered translation and bilingual reading, preserving the original structure and styling where possible.
2. How does WhaleRead's privacy compare to using ChatGPT or Google Translate for books? WhaleRead provides superior data privacy as it never sends your book text to a third-party cloud by default. Translation runs via an on-device 7B AI model or a self-hosted 30B model on your private server, whereas ChatGPT and Google Translate require uploading your content to their servers.
3. Can I use my own AI model with WhaleRead? Yes, WhaleRead is designed for self-hosted AI model integration. You can configure it to use a private 30B parameter model (or similar) running on your own hardware through compatible local inference servers, giving you control over both the data and the model.
4. Does WhaleRead automatically rewrite my original book files? No. WhaleRead employs a human-in-the-loop editing principle. All AI suggestions are drafts. You must use the evidence-first review panel to inspect, edit, and explicitly confirm any change before it is applied to your reading copy, keeping your original file intact.
5. What is the difference between the on-device 7B and self-hosted 30B model quality? Based on the product's tests, the on-device 7B model provides fast, private translations with generally correct grammar. The self-hosted 30B model demonstrates superior performance in preserving idioms, narrative tone, and cultural nuance, offering translation quality comparable to larger cloud models while remaining on your private infrastructure.