MinimalAgents: A lightweight framework for tool-augmented LLM systems
Abhinav Tripathi, Aman Dogra, Prabal Sharma, Sambit Satpathy, Mohit Chowdhary, Dhirendra Kumar Shukla · 2025
Objective: To design a lightweight and modular framework, called Minimal Agents, for building tool-augmented Large Language Model (LLM) systems that reduce complexity and abstraction overhead present in existing solutions. Method: The framework integrates essential tools such as code execution, web search, database querying, and document generation. It follows a minimalist design philosophy to maintain modularity and simplicity, avoiding the layered complexity seen in platforms like Lang Chain, Crew AI, and AutoGen. Findings: Experimental comparisons show that Minimal Agents deliver performance comparable to more complex frameworks, while being significantly easier to understand, extend, and deploy. Novelty: The framework emphasizes a trade-off between capability and complexity, providing a clean and accessible foundation for programmers and researchers to construct AI assistants with minimal abstraction, enabling faster iteration and deployment.