LLM Service

The Text+ LLM Service enables researchers to use Large Language Models (LLMs) for working with text- and language-based research data. Operated by GWDG, the service provides access to open-source language models within the Text+ research infrastructure, allowing AI-supported exploration, analysis, and retrieval of research data while keeping research data within a trusted scientific environment.

GWDG currently provides access to the open-source language models LLaMA (Meta), Mixtral, Qwen, and Codestral. The service supports the development and testing of AI-assisted research use cases and provides a foundation for integrating language models into Text+ services and research workflows.

One of the first applications built on the LLM Service is the Text+ AI Assistant, which enables natural-language access to the lexical resources available through Text+. Instead of learning the formal query language LexCQL, researchers can simply ask questions in natural language, for example “What is the opposite of sweet?” or “Search for the word ‘Lauf’ as a noun.” The assistant automatically translates these requests into valid LexCQL queries and searches the available lexical resources through the Federated Content Search (FCS).

The LLM Service combines the flexibility of large language models with reliable software components to ensure transparent and reproducible results. Rather than sending AI-generated queries directly to the search infrastructure, the service first validates them for correct syntax and checks whether the requested search features are supported by the selected resources. Where possible, unsupported query elements are automatically adapted before the search is executed. This approach allows researchers to benefit from AI-supported search while maintaining compatibility with the underlying research infrastructure.

The service supports multilingual user requests and provides evidence-based answers together with the generated search queries and linked source references, allowing researchers to understand how results were obtained. By lowering the barrier to accessing complex lexical resources, the LLM Service makes advanced search functionality accessible to a broader research community while providing a foundation for future AI-powered services within Text+.

Beyond natural-language search, the LLM Service provides a platform for developing and testing AI-assisted research applications within the Text+ infrastructure. The Text+ AI Assistant also supports Retrieval-Augmented Generation (RAG), allowing researchers to query and extract information from their own documents using large language models.