Agents101 is an educational framework I built for exploring LLM agent design patterns from scratch. It covers the core mechanisms that make agents work, such as tool use and planning.
The framework implements the same patterns across SmolAgents, LlamaIndex, and LangGraph to showcase and compare how each library approaches agent orchestration.
It can serve as a practical companion for engineers who want to understand agents by building them from scratch, not just calling them.
Additional resources
- Hugging Face AI Agents Course — the course this repo grew out of; start here if agents are new to you
- Agents (Chip Huyen) — clear conceptual overview of planning, tool use, and failure modes
- ReAct: Synergizing Reasoning and Acting in Language Models — the paper behind the reasoning + tool-use loop most agents still follow
- smolagents documentation — lightweight code agents with a small API surface
- LangGraph tutorials — stateful, graph-based agent workflows
- LlamaIndex Workflows — RAG-heavy agents and multi-step pipelines