therefore essential for effective PG management. However, existing approaches remain limited. They often (i) assume complete or consistent label annotations, (ii) rely solely on label-based grouping, (iii) cannot handle heterogeneous or noisy data, or (iv) only extract node types without modeling relationships or constraints. Systems such as SchemI [7] and GMMSchema [3] and DiscoPG [2] perform well when data is fully annotated and clean, but degrade substantially under missing properties, incomplete labeling, or structural inconsistencies—conditions that are typical in real-world PGs. To address these challenges, we present PG-HIVE, a hybrid, incremental, and adaptive system for schema discovery on property graphs. PG-HIVE employs a two-stage inference strategy that combines semantic embeddings of labels with structural similarity derived from properties, clustering nodes and edges using Locality-Sensitive Hashing (LSH)—either via Euclidean LSH or MinHash. This hybrid representation allows PG-HIVE to infer schema elements even when labels are partially missing or entirely absent. A second merging phase refines the discovered clusters into coherent types based on labels and –if unlabeled– high-similarity structural patterns. PG-HIVE then infers mandatory vs. optional properties, property datatypes, and relationship cardinalities, and exports the resulting schema in widely compatible formats, including PG-Schema and XSD. PG-HIVE also has an incremental module, which allows incremental updates in the schema as new of data arrives without recomputing the dataset from scratch. Through extensive experiments on eight real and synthetic datasets, including heterogeneous, multi-labeled, and noisy graphs, PG-HIVE has been shown to consistently outperform state-of-the-art solutions [10]. It improves schema discovery accuracy by up to 65% for node types and 40% for edge types, and runs up to 1.95× faster than prior work for full schema inference. Crucially, PG-HIVE maintains accuracy even under extreme conditions, such as 0% label availability and up to 40% missing properties, where competing systems fail. In this demonstration, we showcase the full interactive experience of PG-HIVE. Users can load a property graph from a backend store (e.g., Neo4j), explore clustering options (manual or adaptive), inspect intermediate and final schema elements, visualize discovered node and edge types, inspect property constraints, and compare schema extraction results under different parameter settings or noise levels, and baseline approaches. The demo highlights PG-HIVE’s ability to (i) discover types in challenging scenarios, (ii) incrementally update schemas, and (iii) give users full control and transparency over the inference process Property graphs are increasingly used to model complex and highly interconnected data across a wide range of domains. However, their schema-free nature continues to hinder understanding, integration, and efficient management. In this demo, we present PG-HIVE, an interactive, end-to-end framework for automatic schema discovery in property graphs. PG-HIVE unveils latent node and edge types, infers property datatypes and cardinalities, and supports both user-driven and fully adaptive clustering through a hybrid, incremental pipeline that avoids costly recomputation as new data arrives. The demonstration showcases a lightweight, intuitive web interface connected to a graph storage backend, enabling users to explore discovered schemata in real time through rich visualizations and customizable discovery settings. Beyond usability, PG-HIVE delivers state-of-the-art performance, improving schema-discovery accuracy by up to 65% for nodes and 40% for edges and achieving up to 1.95× faster execution compared to existing approaches. Designed for rapid deployment and experimentation, PG-HIVE empowers practitioners to perform schema discovery interactively, at scale, and on the fly, unlocking a deeper understanding of property graph structure.