Exploring Knowledge Graphs (KGs, in short) to discover facts and links is tedious even for experts with knowledge of SPARQL due to their unfamiliarity with the structure and labels of entities, classes and relations. Some KG applications require finding the connections between groups of nodes, even if users ignore the shape of these connections. However, SPARQL only allows checking if paths exist, not returning them. A recent property graph query language, GPML, allows also returning connecting paths, but not connections between three or more nodes. We propose to demonstrate RelSearch, a system supporting extended SPARQL queries, featuring standard Basic Graph Patterns (BGPs) as well as novel Connecting Tree Patterns (CTPs); each CTP requests the connections (paths, or trees) between nodes bound to variables. RelSearch evaluates such extended queries using novel algorithms which, unlike prior keyword search methods, return connections regardless of the edge directions and are independent of how we measure the quality (score) of each connection. We will demonstrate RelSearch's expressivity and efficiency using a variety of RDF graphs, user-selected score functions, and search exploration orders.