Command and Conquer: How we let an LLM control our Software

In this session, we’ll share our journey of building a system where a Large Language Model can control almost any part of our software, guided by a Chat and Generative UI. We’ll explore the architecture we created to support any LLM, allowing users to choose from models from OpenAI, Google, Anthropic or Mistral. We will share the many practical lessons we learned: what works, what fails spectacularly, and how different LLMs excel in different areas. You’ll see a practical demonstration of how we add new commands to the model’s capabilities, iterate on functionality, and, most importantly, how we architected the system to always keep the “human in control”. This isn’t just another AI wrapper and not an MCP; we’ll discuss unique security challenges, how we solved them, and how our approach enables compliance and user control.