Introduction

The Local AI Translator is a powerful, fully offline AI translation application that runs large language models (LLMs) directly on the user’s own computer, with no internet connection required and no data sent to an external service. Built on top of Ollama, it provides a rich graphical interface and professional translation environment that goes far beyond simple command-line interaction, making advanced AI translation accessible to everyday users. The ability to deploy powerful LLMs entirely offline makes it an ideal solution for security-conscious organizations that cannot risk sensitive content being transmitted to third-party servers.

The Local AI Translator also has particular relevance for the Global South, where access to expensive cloud-based AI translation services is often limited by cost, unreliable internet connectivity, and foreign currency constraints. With no subscription fees or API costs, it puts capable AI translation tools within reach of researchers, educators, developers, and small businesses in countries where paying for services priced in US dollars represents a significant financial burden. Once a model has been downloaded, the application works completely offline — a major advantage in regions where internet access is intermittent or bandwidth is costly — and directly addresses the data sovereignty concerns that are increasingly important to governments and institutions wary of sensitive information being processed on servers in foreign jurisdictions.

The Local AI Translator runs offline on any Windows 11 PC with at least 16 GB of RAM, requiring no specialist hardware and no GPU.  Bespoke Linux versions are available on request.

Main features

The main features of the application are identified by the row of buttons in the figure below: Model Manager, Prompt Manager, RAG Manager, Backtranslate, Batch Translation and Tools. The Model Manager offers functions for downloading (“pulling”), changing and deleting LLMs from within the GUI, direct access to the Ollama Model Library and a Model Evaluator.

The Model Evaluator gives  the user a way to assess the performance of a series of LLMs for specified language pairs by reference to various commonly used metrics.   Given the growing number of LLMs which are claimed to translate between numerous languages, the evaluator is a valuable tool for users to assess which LLMs are likely to generate a “usable” translation for their use cases.  The module provides scores for each translation but also allows users  to save each translation with the name of the generating model.

The Prompt Manager provides a space where the user can import previously used prompts or create custom prompts. These can range from brief terminology or style guidelines to comprehensive instructions or even extensive glossaries.

The RAG Manager allows the user to deploy RAG (Retrieval Augmented Generation) in three ways: importing a flat list of glossary terms, building a knowledge graph from a source text and an existing translation memory and importing a knowledge graph in one of the commonly used formats (JSON, ttl, gml).  By checking and unchecking the “Use RAG” checkbox the user can switch between translation with and without RAG.  Using Knowledge Graph RAG can be a sophisticated way of injecting specialized knowledge into the prompt to the LLM, but adding a short glossary at the head of the text to be translated may be an effective approach if a large LLM is deployed.

The “Translate File” function enables the user to translate text files at three levels: sentence level, paragraph level and document level.  Working at paragraph or document level gives the LLM the broader context needed to deploy appropriate terminology and style.  Document level processing is essential when a large glossary or knowledge graph is injected into the prompt.  Besides translating text files, the Local AI Translator can also process MS Office files,  Libre Office files and pdf files.   The “Batch Translation” module handles the translation of  number of selected files in any mix of  formats and any combination of languages.

The “Tools” button provides the user with ten utilities (Model Supported Language Checker, AI Chat box, Image Analyzer(Intelligent OCR), Bitext Creator, XLIFF Generator, Transcriber, Rewriter, Proofreading & Correction tool, Terminology Extractor, Summarizer, Output Screen Reader.   All these tools are intended to facilitate or assist the primary translation task. For example, the AI Chat box can be used to do ad-hoc research or generate a glossary in a specific field which can then be used for translation by the Prompt Manager or the RAG Manager.  The Image Analyzer can not only extract text contained in an image but can reason about the image itself.

The latest feature to be introduced in the application is “Interactive Mode”.  Selecting  “Interactive Mode” brings up a grid which will be familiar to users of CAT tools like Trados and MemoQ.   The grid allows for sentence-level or paragraph level segmentation as specified by the user. The source language, target language and model are set prior to entering Interactive mode, but the LLM can be changed during the translation process.  The grid design allows the user to intervene in the translation process by editing a segment or entering their own translation in any segment box.  The current state of the Interactive Project can be saved and re-loaded later.  The resulting translation can then be saved with the source text as an XLIFF file, a DOCX file, a bilingual DOCX file, an ODT file , a TSV file or a simple text file.

Please note that all sales are final. Due to the nature of digital products, once the software has been downloaded, we are unable to offer refunds under any circumstances. By completing your purchase, you acknowledge and agree to this no-refund policy.