emcie-co/parlant ? reverse-engineered prompt

Build me a Python framework for creating advanced AI agents, specifically for customer-facing roles like support or sales. I'm finding that when I use a single huge system prompt for my chatbot, it gets confused and starts ignoring rules as I add more complexity.

I want a more robust system that manages the conversational context intelligently. Instead of giving the LLM all the rules at once, the framework should figure out which instructions, knowledge, and tools are relevant for each specific turn in the conversation. For example, if a customer seems confused, the agent should simplify its language. But if the customer uses expert financial terms, it should recognize that and give a more technical, in-depth response.

The goal is to define all this behavior in Python code, not prompts. It needs to be super reliable and auditable, since it'll be used in a high-stakes industry like banking. Ultimately, I want a system where adding more rules makes the agent smarter and more aligned, not more chaotic.