Launching today

WhaleRead

Launching today

Locally translate TXT, Markdown, and EPUB

13 followers

Locally translate TXT, Markdown, and EPUB

13 followers

Visit website

Translation

WhaleRead is a local-first macOS reader that translates TXT, Markdown, and EPUB with on-device 7B or self-hosted 30B private models, preserves bilingual reading, and adds human-confirmed review plus Ask AI—while keeping your library under your control.

Free

Launch tags:Languages•Artificial Intelligence•Books

Launch Team

OpenAI

GPT-6 Astra Challenge

What became possible in your product with Astra that was not practical before?

WhaleRead

Maker

WhaleRead began as a useful Chinese-first reader with fragile language coupling and an unfinished release path. Astra raised the ambition from “add an English UI” to a verifiable, worldwide private-reading product. It audited the architecture, separated interface language, translation direction, and AI answer language into explicit boundaries, designed evidence gates for delegated implementation, and independently replayed the packaged app. That enabled 18 translation directions with native language names, live Chinese/English switching, stable answer-language behavior, bilingual TXT/Markdown/EPUB reading, human-confirmed review, and Ask AI with visible context. Astra also helped turn real on-device 7B and private 30B outputs—including their mistakes—into an honest demo using only original synthetic text. The result is a safer product and a repeatable way for a non-programmer founder to coordinate multiple models without losing the product’s architectural intent.

Report

3h ago

WhaleRead

Maker

📌

Hi Product Hunt — I built WhaleRead because translating a book often destroys the experience of reading it. WhaleRead keeps the whole journey in one calm macOS workspace: import a local TXT, Markdown, or EPUB, translate it, read source and translation side by side, flag an awkward passage, review a proposed correction, and ask AI about the page without moving the whole library into a consumer cloud. For this launch I tested the same five original passages with two private model paths. The on-device 7B keeps text on the Mac and works well for everyday reading. A self-hosted 30B private model handled idioms and narrative voice more consistently, with quality that feels close to cloud-scale models while staying on infrastructure I control. I also kept the failures visible. In one English-to-Chinese passage, the 7B translated “kept her cards close” as literally holding cards to her chest. WhaleRead’s review flow surfaced the source, explanation, suggested revision, and a final human-confirmed fix. In a French test, even the 30B produced a title/body gender inconsistency. That is why review is part of the product, not a cleanup step hidden from the reader. WhaleRead 1.18 has English and Chinese interfaces, native language names, bilingual reading, EPUB preservation, resumable translation, human-confirmed review, and page-based Ask AI. GPT-6 Astra helped audit the architecture, harden language boundaries, and turn the release into a verifiable demo. This is a working macOS private beta. I would especially love feedback from multilingual readers, translators, and people who care about owning their reading data: which language pair or difficult expression should I test next?

Report

7h ago