AI News

TrustGraph presents itself as an open-source context orchestration layer for giving AI agents more accurate and explainable access to organisational knowledge. The article argues that conventional Retrieval Augmented Generation (RAG), which retrieves text chunks through vector similarity, can overwhelm language models with loosely relevant context, provide little control over what the model sees, and fail to answer relationship-specific questions. It also says vector retrieval cannot reliably distinguish contextual uses of terms, using the “Who’s on First?” wordplay as an example, and does not show which sources or reasoning steps produced an answer. TrustGraph instead extracts entities, relationships, and facts from documents and stores them with vector embeddings in a context hypergraph. Unlike a simple pairwise graph, the hypergraph can connect several entities and metadata points to a single event or conceptual unit. When an agent asks a question, the system retrieves structured context, such as a player connected to the FirstBase position, rather than relying only on probabilistic text similarity. The article describes claimed benefits including more focused context, relationship-aware retrieval, user-defined ontologies through “Bring Your Own Ontology,” and control over the knowledge presented to the model. It also says answers can be traced to hypergraph nodes, source documents, and reasoning steps, with the provenance described as cryptographically verifiable. The page positions TrustGraph as infrastructure for AI systems that need to reason over organisational knowledge and show how they reached an answer.