Getting started
Glossarion is free and open source. There is no account, no store and no subscription: every build comes from the GitHub releases page. You bring the AI — your own API key, a subscription login, or a model running on your own machine.
Short version: download the build for your device from the latest release, open it, type a model name (for example gemini-3.1-flash-lite with a free Google AI Studio key, or ollamapull/qwen3.5:9b for a local model), choose a book, and press Run.
Which download do I need?
Every release ships these files. Version numbers in the names change with each release.
| Device | File | Notes |
|---|---|---|
| Windows | Glossarion.vX.exe ⭐ | Standard build — recommended for novels. Includes the EPUB library/reader and the SDLXLIFF editor. No manga translator. |
N_Glossarion_NoCuda.vX.exe | Everything, including the manga translator and the optional NER name check. Best for manga. No NVIDIA GPU needed. | |
L_Glossarion_Lite.vX.exe | Smaller: drops the EPUB reader, the SDLXLIFF notepad mode and browser-based routes. | |
L_Glossarion_TurboLite.vX.exe | Smallest and fastest to start: also drops the Vertex AI SDK and PDF generation. | |
| macOS | …_MAC.dmg / …_MAC_Intel.dmg | Apple Silicon (M1 and newer) or Intel Macs. NoCuda includes the manga translator; Lite does not. |
| Linux | Z_Glossarion_vX_Linux.zip | Standard build. NoCuda (manga) and TurboLite variants are also published. |
| Android | …_Android_arm64-v8a.apk | Almost every phone and tablet. Use x86_64 only for emulators and Chromebooks. |
| iOS / iPadOS | …_iOS_unsigned.ipa | Unsigned — install with a sideloading tool (see below). |
Not sure? On Windows take the standard build; switch to NoCuda only when you want to translate manga.
Windows
- Download the
.exeyou picked from the latest release. - Double-click it. There is nothing to install — the program is a single file.
- The first start shows a splash screen and can take 10–60 seconds while files unpack. Later starts are faster.
"Windows protected your PC"? The builds are not code-signed, so SmartScreen may warn about an unrecognised app. Choose More info → Run anyway. If your antivirus is unsure, you can always build from source instead.
macOS
- Download the
.dmgfor your Mac:_MACfor Apple Silicon,_MAC_Intelfor Intel. - Open the disk image and drag Glossarion into Applications.
- The app is not notarised, so the first time macOS may refuse to open it. Right-click the app and choose Open, or go to System Settings → Privacy & Security and click Open Anyway.
Linux
- Download the
_Linux.zipand unzip it. It contains a single executable. - Make it executable and run it:
chmod +x Glossarion* ./Glossarion* - The build uses Qt. If it complains about the
xcbplatform plugin, install the XCB libraries — on Debian/Ubuntu:sudo apt install libxcb-cursor0.
Android
- On your phone, download the
_Android_arm64-v8a.apkfrom the latest release. - Open it. Android asks you to allow installs from your browser or file manager the first time — allow it, then tap Install.
- To update, install the newer APK over the old one; your books, keys and settings stay.
The mobile app is chat-first: talk to the model, drop in a book, and long jobs keep running in the background with a notification. It can also run small GGUF models on the phone itself — no server and no internet once the model is downloaded. A 4B model wants about 4 GB of free RAM.
iOS & iPadOS
Glossarion is not on the App Store, so the iOS build is an unsigned IPA. Install it with any sideloading tool that signs the app with your own Apple ID, for example SideStore, AltStore or Sideloadly:
- Download
_iOS_unsigned.ipafrom the latest release. - Open it with your sideloading tool and let it sign and install the app.
- With a free Apple ID, apps signed this way must be refreshed every 7 days; your sideloading tool does this for you.
Long books on iPhone: iOS pauses a translation about 30 seconds after you leave the app. Keep Glossarion open while a long job runs, or translate long books on Android or desktop.
On-device models on iOS: sideloaded apps cannot ask for Apple's increased-memory entitlement, so iOS lets a local model use at most about half of the device's RAM. Small models (0.8B–4B) are the sweet spot.
Run from source
Prefer Python? Glossarion runs from source on Windows, macOS and Linux with Python 3.10 or newer.
git clone https://github.com/Shirochi-stack/Glossarion.git
cd Glossarion
pip install -r requirements.txt # macOS: requirements-macos.txt
cd src
python translator_gui.py
On Windows you can also double-click src/launch_Glossarion.bat.
Pick a model
Glossarion does not sell AI access, so you choose where translations come from. The Model box is a free-text field: type the model name, including its prefix, and Glossarion routes the request.
| You want… | Type in Model | What you need |
|---|---|---|
| Free, on your own PC | ollamapull/qwen3.5:9b | Nothing — Glossarion installs Ollama and downloads the model for you. |
| Free, on your phone | Model manager › On-device | A phone with enough free RAM for the model (Qwen3.5 0.8B–9B presets). Mobile can also use Ollama or LM Studio running on a PC on your network. |
| Free cloud tier | gemini-3.1-flash-lite | A free Google AI Studio key (about 500 requests a day on Flash Lite). |
| Free models on OpenRouter | or/…:free | A free OpenRouter key. |
| Use a subscription you have | authgpt/…, authgrok/… | Sign in with the 🔐 button next to the model field. |
| Top quality | gpt-5.4, claude-sonnet-4-6, gemini-3.1-pro-preview, deepseek-chat… | An API key from that provider. You pay the provider directly, per use. |
| Your own server | ollama/…, lmstudio/… or a custom prefix | Any OpenAI-compatible endpoint. |
The full list of 40+ providers and every free route is in the user guide.
Your first translation (desktop)
- Choose a file — an EPUB, TXT, PDF, subtitle file, image or manga archive.
- Pick a profile for the source language (Korean, Japanese, Chinese and more). The profile holds the translation prompt; edit it freely.
- Type a model and paste its API key (skip the key for local models and logins).
- Set a chapter range to try a couple of chapters first.
- Turn on the glossary so names stay the same in every chapter (see Glossary).
- Press Run. Progress, the API watchdog and logs update live; Stop and resume any time.
- Open the output — a translated EPUB is built for you. Read it in the built-in library, then run a QA Scan to catch untranslated or duplicated passages.
On mobile the flow starts in the chat: tap + to attach a book from your files or Library, check the plan card, and start the job. It keeps running in the background with a notification, and the finished book lands in your Library. To translate a few lines, just paste them into the chat.