raga-ai-hub/RagaAI-Catalyst ? reverse-engineered prompt

Build me a Python SDK for monitoring and evaluating LLM and agent apps.

I want to be able to create projects, manage datasets, run evaluations on saved data, and record traces from my app so I can see what the model, tools, and agent are doing step by step. Please include support for tracing normal LLM calls and multi agent workflows, plus a simple way to start and stop tracing in code.

Also add prompt management, synthetic data generation, and basic guardrail and red teaming features since this project is meant to help with safer LLM apps. It should feel like a real SDK, easy to import and use from Python, with authentication through access and secret keys and a base URL. If you need current library or API details, look them up online and make the implementation fit the latest docs.