Knowledge File

A Knowledge File is a resource that provides additional context to Large Language Models (LLMs) in Hellixia, BayesiaLab’s LLM-powered assistant. Its role is to supply domain‑specific information, beyond what the LLM already knows from its training—so that Hellixia can generate more accurate, grounded Semantic Networks, Knowledge Graphs, and causal models.

Hellixia supports a wide range of file formats for Knowledge Files: plain text (txt), PDF, Word, Excel, RTF, HTML, XML, PowerPoint, and several audio formats (mp3, pcm, wav, aac, opus, flac). This breadth means you can feed both written documents (articles, book chapters, reports) and spoken content (recordings, lectures) as context.

In practice, a Knowledge File can be a single long narrative, like a scientific article or a book chapter—or a text compiled by a Retrieval-Augmented Generation (RAG) tool, which automatically selects the most relevant excerpts from an embeddings database for the current query. For instance, the Knowledge File is used in Document Analysis, where Hellixia feeds the file’s content to the LLM to extract concepts and relations that reflect the document’s specific knowledge.

Think of a Knowledge File as a memory supplement for the LLM: even though the model contains vast general knowledge, the Knowledge File lets you inject precise, up‑to‑date, or proprietary information about your problem domain, making Hellixia’s outputs more relevant and trustworthy.