Create a lightweight Python wrapper that standardizes agentic workflow execution across disparate inference engines like M2.7, Llama, and Mistral. Focus on reliability in multi-step task chaining.