Getting Started

MetalGlot is a desktop app for local AI translation. It runs a Docker-based translation backend on your own machine, then gives you dedicated workspaces for resource files, raw text, subtitles, Markdown, and image translation.

This guide walks through the fastest setup path: choose the right Docker image, create a translation container, assign a model folder, download a GGUF model, load it into memory, and open the workspace that matches your source format.

What You Need First

• Docker Desktop installed.
• Docker Desktop running before you try to create or start a translation container.
• Enough disk space for at least one Docker image and one GGUF model.

When the backend is running, the app talks to a local service at http://127.0.0.1:8080. That service stays on your machine rather than in a remote cloud account.

Safe default

If you are not sure which container image or model to choose for a first run, start with the CPU image and a 4B or 12B model.

Quick Start

1. Open Setup Container and download the CPU or CUDA image that matches your hardware.
2. Create a container and choose a model folder for that container.
3. Start the container from the home page toolbar.
4. Open Manage AI models and download a GGUF model into the selected container’s model folder.
5. Load the model into memory.
6. Open the translation page that matches your file type.

If you want to use the app in one of its 32 localized interface languages, switch the UI language from the home page language selector before you continue. See Interface Language Selector.

Related Guides

• Interface Language Selector.
• Docker Images and Containers.
• Translation Pages and Workspaces.
• Inference Settings.
• GGUF Models and Downloads.
• Translation Wizard and Validation Warnings.

Local Backend Note

For text translation tasks, MetalGlot sends POST requests to the local backend with the loaded model, source language, target language, the text content, and any enabled inference settings. Depending on the language pair, the app uses either the standard text endpoint at /translate or the experimental text endpoint at /experimental_translation for partially supported language combinations.

For image translation, use the Image Contents workspace with a supported vision-enabled model.