Dodari 2 translates whole books locally or via your existing AI subscription

Dodari 2 is a multilingual translator built for entire documents rather than short snippets. The README describes it as the successor to Dodari 1, released in March 2024, and it accepts EPUB, PDF, and TXT files across Korean, English, Japanese, Chinese, French, Italian, Dutch, Danish, Swedish, Norwegian, Arabic, and Persian. The workflow is aimed at readers and small teams who need a finished book file, not a browser copy-paste session. It detects language and book genre automatically, lets the user choose a translation tone or style, and extracts key terms such as names into a glossary to apply consistently. Each run produces two outputs: a translation-only file and a version that keeps the original alongside the translation for sentence-by-sentence comparison. That second file is the most useful detail for review, because it makes spot-checking terminology and omissions practical without reopening the source separately. Two ways to run it Dodari 2 can use an existing Claude or ChatGPT subscription through the official CLI, or run local Gemma models. For the subscription path, the README lists Codex or Claude Code as the required app, with ChatGPT Plus or Claude Pro as the minimum plan. The Claude engine defaults to the Haiku model named in the README at maximum reasoning effort, and the model and effort can be changed in the UI or in a local dodari_config.json. The local path lists Gemma4 e4b 8bit as the minimum tier and Gemma4 31b 4bit as the recommended tier, with corresponding storage, memory, and GPU requirements that vary sharply by platform. Subscription CLI: no per-token API setup in the app itself, but it depends on an external CLI and a paid plan remaining eligible. Local Gemma: no subscription needed for translation, but first-run model downloads are large and the recommended Windows tier calls for 24 GB VRAM or more. Local UI: scripts for Windows, macOS, and Ubuntu launch the app at http://127.0.0.1:7860,