danterolle/loqi

Loqi is a command-line translation tool that runs entirely on your computer, sending no text to Google, DeepL, or any third-party service. You give it text, a file, or piped input, and it translates using a local AI model. The author built it out of a desire to stop sending personal and business documents to external translation services, even those with opt-out settings and privacy policies. The tool supports three different backends. The default is Ollama, which can download and run a language model locally and is the easiest option to start with. The second backend is llama.cpp, which connects to a locally running model server you point at a GGUF model file. The third is Argos Translate, a rule-based translation engine that works fully offline with no GPU, installs itself automatically in a Python environment, and handles straightforward sentences faster than a language model but with less context awareness. There are three ways to use Loqi. Interactive mode opens a terminal interface where you type text and see translations update as you type. One-shot mode runs a single translation from the command line and returns the result, which works well for scripting and piping into other tools. Batch mode processes JSON files, plain text files, or markdown documents. The tool is written in Go and installed with a single go install command or by building from source. Configuration lives in a YAML file, and all settings can be overridden with command-line flags. The translation quality depends on the model you choose, smaller models make mistakes, and the README recommends treating output as a draft to review rather than a finished translation. Loqi is licensed under Apache 2.0.