Dragomand

Dragomand is a per-user Linux service that gives desktop applications a shared way to translate text locally. It runs Mozilla's Bergamot and Marian translation technology, the same foundation used by Firefox Translations, and exposes it through a D-Bus interface. An editor, desktop widget, office extension, shell command, or accessibility tool can call the same service instead of bundling a separate translation engine and another copy of every model. Once a requested language model has been downloaded, the text being translated stays on the computer.

The product addresses an awkward gap in Linux translation tools. A standalone translator may work well for one task, while translation embedded in a browser is useful only inside that browser. Developers who want translation in another application otherwise need to ship an engine, manage model downloads, and absorb the memory cost themselves. Dragomand centralizes those jobs. Every compatible client talks to one daemon, installed models live in one user store, and a model that is already loaded can serve several applications. That shared architecture is the main reason to consider it over another isolated desktop translator.

Day to day use can be as direct as running dragomanctl with a source language, a target language, and some text. The first request for a pair downloads its model from Mozilla, verifies it with SHA-256, starts the daemon through D-Bus activation, and returns the translation. Later requests for that pair work offline. The command line client can also read one segment per line from standard input, write translated output to a file, and mark large work as batch traffic so interactive requests can move ahead. Users can list installed and available pairs, prefetch models before travel, check memory use and queued jobs, update models, or remove pairs they no longer need.

The daemon is designed to stay out of the way when it isn't needed. D-Bus starts it on demand, and it exits after an idle period rather than occupying resources permanently. Dragomand can keep recently used models warm, evict them with a least recently used policy, and respond to memory pressure within a configurable budget. Model updates happen only when the user asks or enables the optional weekly systemd timer. The available model catalog is cached, so it can still be inspected without a connection. Network access is for obtaining models and updates, not for sending translation text to a remote service. Whole document requests, language detection, sentence alignment, and model quality information are available through the shared API, so clients can do more than submit a single short phrase.

Dragomand is most useful to Linux users who want private translation across several desktop workflows and to application developers who would rather call a stable system service than embed machine translation. The project documents integrations for KRunner and GNOME Shell search, LibreOffice, and the KTextEditor component used by Kate, KWrite, and KDevelop. Its companion KDE tools include the Krakoman graphical translator, a System Settings module, and a Plasma widget. Sandboxed Flatpak applications need explicit permission to talk to the D-Bus name, so integration is not completely automatic in every packaging environment.

Its maturity and language coverage need to be read carefully. Version 0.2.0 includes the daemon, its first D-Bus API, document translation, language detection, sentence alignment, live settings, model management, and several desktop integrations. Bulgarian and English are the first pair working through the complete path. The broader goal covers every pair Mozilla publishes, but buyers should not read that roadmap as a promise that every pair already has the same end to end support. It is also infrastructure rather than a universal polished app on its own, although companion interfaces make it approachable for KDE users.

Dragomand is free and open-source software under GPL version 3 or later. Packages are published through the openSUSE Build Service for current openSUSE, Fedora, Debian, Ubuntu, and Arch Linux targets listed in the installation guide, and experienced users can build the C++ engine and Rust workspace from source. There is no hosted subscription or metered translation plan. The practical costs are local disk space, model memory, and the setup needed for the distribution and desktop clients a user chooses. For people comfortable with that Linux native model, it offers a focused way to add offline translation once and make it available across the desktop.