Neuro-symbolic (NeSy) AI systems are increasingly used in domains such as healthcare and finance, yet their architectures are often described only in prose or diagrams, making them difficult to compare, reproduce, or integrate. Existing modelling tools, such as UML or ML workflow editors, do not capture the semantics of NeSy components or enforce syntactic and semantic constraints on architectural validity. To address this gap, we provide a semantic, valid, and interoperable representation of NeSy architectures and present Tool4Boxology, a toolbox for constructing, validating, and analysing such architectures. The resource provides (i) a graphical editor for architecture diagrams with syntax validation, (ii) an ontology that describes architectural components and their interconnections, (iii) a pipeline that uses the RDF Mapping Language to generate ontology-compliant knowledge graphs from diagrams, (iv) SHACL constraints that enforce semantic coherence conditions, and (v) querying and validation mechanisms on these knowledge graphs for pattern detection, correctness checking, and reuse. We demonstrate the utility of Tool4Boxology by modelling a corpus of 61 NeSy systems from the biomedical literature and 7 from other domains, enforcing correctness constraints on these models, and querying them for recurring sub-patterns. Tool4Boxology offers researchers and practitioners a transparent, comparable, and FAIR-aligned foundation for documenting and analysing NeSy AI systems. The application, source code, ontology, and a corpus of neuro-symbolic architectures are available under a permissive license, along with an instruction video. A Jupyter Notebook provides runnable examples of KG exploration and analysis.