// the find
huggingface/agents-course
This repository contains the Hugging Face Agents Course.
A structured, free course from Hugging Face covering AI agent fundamentals through practical implementation with smolagents, LangGraph, and LlamaIndex. It culminates in a GAIA benchmark evaluation where students submit agents for automated scoring. Aimed at Python developers who know LLMs exist but haven't built anything agentic yet.
Unit 1 builds the mental model correctly — ReAct loop, tool calls, special tokens — before touching any framework, which is the right order. The three-framework coverage in Unit 2 is genuinely useful: you see the same problem solved with smolagents (minimal), LangGraph (stateful graphs), and LlamaIndex (data-heavy), so you can make an informed choice. The bonus observability unit covering tracing and evaluation is rare in courses like this and addresses the part most tutorials skip entirely. Active multilingual community contribution (EN, ES, FR, KO, RU) with translation agreements in place means the material isn't rotting.
The repo is almost entirely MDX prose and Jupyter notebooks — there's no reference implementation you can clone and run as a production-grade agent, so the gap between 'I finished Unit 4' and 'I can ship something' is still large. GAIA as the final benchmark measures general assistant capability, not domain-specific agentic reasoning, so high GAIA scores don't necessarily mean you've learned to build useful agents for real tasks. The smolagents framework featured prominently is Hugging Face's own library, which creates an obvious conflict of interest in how it's presented relative to LangGraph and LlamaIndex. Course content is frozen at a point when the agentic tooling landscape was moving fast — what's taught about LangGraph in particular may already lag the current API.