grounded directly in historical documents, enabling users to verify claims against the archival record. We present a system that answers natural-language questions using digitized historical newspapers. Our prototype leverages the scanned newspaper archives of the Canton of Vaud in Switzerland to support evidence-based exploration of regional history. We formulate the problem as an embedding-based retrieval augmented generation (RAG) task: each scanned article is transcribed through OCR, encoded into a high-dimensional semantic embedding space using a large language model, and indexed for efficient similarity search. User queries are embedded in the same space, and the top-k nearest articles are retrieved as candidate evidence from which the system synthesizes an answer. To promote transparency and verifiability, the original retrieved articles are always displayed alongside the generated response. Users can browse these primary sources either on a temporal axis, visualizing the evolution of events over time, or on a geographical map that highlights spatial relationships relevant to the query. To ensure both high performance and correctness, we employ a metric-tree index structure that guarantees no false negatives in k-nearestneighbor retrieval while supporting low-latency exploration at scale. By integrating data management techniques with RAG, our system offers an interpretable and provenance-aware alternative to modern information systems, enabling fact checking directly against historical documents rather than relying solely on model-generated text. This demonstration presents HistoricRAG, an iPad-based application that lets users ask questions about historical or other events and immediately inspect the documentary evidence behind each answer. The system couples semantic retrieval over a large collection of OCR-processed Swiss newspapers with a generation component that summarizes the retrieved passages while preserving their provenance. Every response is accompanied by the original scanned articles, which can be explored along an interactive timeline or viewed through a map that reveals the geographic context of reporting. These features encourage users to crosscheck claims directly against primary sources rather than relying on unverified narratives. By uniting efficient vector search, structured query interpretation, and evidence-centered presentation within a lightweight mobile interface, HistoricRAG offers a practical path toward misinformation-resilient question answering and demonstrates how generative models can be grounded in curated archival material for research, education, and public history.