Model-Based Engineering relies on domain-specific models and diagrams to support the design and analysis of complex systems. Constructing such models remains a time-consuming and error-prone task that requires both deep domain knowledge and great expertise. While Large Language Models (LLMs) offer new opportunities for natural language-driven modeling, they struggle with domain alignment, semantic validation, and reliable formalism generation. In this paper, we introduce an end-to-end agentic approach for diagram generation that combines the reasoning capabilities of LLMs with the native tool-based modeling environment of Sirius Web. Our solution first interprets a high-level user request through a Prompt Interpreter, translating it into structured modeling actions enhanced by domain awareness. Those modeling actions are then processed by an orchestration loop, which delegates the execution to tooling agents. Those agents then leverages Sirius Web internal processes to ensure the generation of semantically valid and syntactically correct diagrams. We evaluate the generation capabilities of our prototype across several domains using different views with specific constraints. The evaluation demonstrates the system’s adaptability and success in generating models under various conditions. However, limitations such as the introduction of irrelevant classes and difficulties with links and containment were observed.