Hugging Face AI Agents Course
by Hugging Face
Learn agents three ways — smolagents, LlamaIndex and LangGraph — and finish on a public benchmark leaderboard.
Overview
The Hugging Face AI Agents Course is a free, self-paced curriculum that carries you from the definition of an agent to a working agent submitted against a public benchmark, and its distinguishing choice is teaching three competing frameworks rather than promoting one. It is organised as Unit 0 Onboarding (accounts, tokens, Discord, notebook setup), Unit 1 Introduction to Agents (the Thought-Action-Observation loop, tools, messages, special tokens and chat templates, and building tools as ordinary Python functions), Unit 2 Frameworks — split into 2.1 the smolagents framework, 2.2 the LlamaIndex framework and 2.3 the LangGraph framework, so the same ideas are implemented three times across three APIs — Unit 3 a use case on agentic RAG, and Unit 4 the final project, where you create, test and certify an agent against a selected benchmark, submit it for automated evaluation, and land on a student leaderboard. Three bonus units cover fine-tuning an LLM for function-calling, agent observability and evaluation, and agents in games using Pokemon. Each chapter is designed for about a week at 3-4 hours, there is no deadline, and everything including certification is free: a fundamentals certificate for completing Unit 1, and a certificate of completion for Unit 1 plus one use-case assignment plus the final challenge. The stated prerequisites are basic Python and a rough understanding of LLMs, with a recap built into Unit 1. Course material is developed in the open at github.com/huggingface/agents-course (30.8k stars, 2.2k forks, Apache-2.0) with community translations, so errata and framework changes land publicly rather than leaving a recorded video quietly wrong.
At a Glance
- Topic
- Agentic
- Level
- Beginner
- Format
- Course
- Cost
- Free
- Duration
- 4 chapters at ~1 week each, 3-4 hours per week (~15-20 hours total), self-paced with no deadline
- Provider
- Hugging Face
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓Explain the Thought-Action-Observation loop and identify where each step tends to fail
- ✓Write tools as ordinary Python functions and expose them to an agent correctly
- ✓Read chat templates and special tokens well enough to debug a broken agent prompt
- ✓Build the same agent three ways in smolagents, LlamaIndex and LangGraph and compare the trade-offs
- ✓Assemble an agentic RAG pipeline that reasons over retrieval rather than just retrieving
- ✓Instrument an agent for observability and evaluate its behaviour, not only its final answer
- ✓Fine-tune an LLM specifically for function-calling in the bonus unit
- ✓Submit a finished agent for automated benchmark evaluation and see it scored on a public leaderboard
Highlights
- •Teaches three frameworks side by side (smolagents, LlamaIndex, LangGraph) instead of locking you into one vendor
- •Ends in automated evaluation against a real benchmark and a public student leaderboard, not a multiple-choice quiz
- •Two certificates — fundamentals after Unit 1, completion after the final project — with no fee at any stage
- •Developed openly on GitHub at 30.8k stars under Apache-2.0, so material tracks framework changes
- •Explicit weekly pacing of 3-4 hours with no deadline, which makes it realistic alongside a full-time job
- •Bonus units cover the parts most courses omit entirely: observability, evaluation and function-calling fine-tunes
Who It’s For
Best For
- ✓Developers with basic Python who have never built an agent before
- ✓Engineers evaluating which agent framework their team should standardise on
- ✓Self-learners who want a credential and a leaderboard result to point at
- ✓Teams running an internal upskilling cohort on a fixed weekly cadence
Prerequisites
- •Basic Python — functions, imports, and the ability to read a traceback
- •A rough working understanding of what an LLM does; Unit 1 includes a recap
- •A free Hugging Face account and an access token
FAQ
What is Hugging Face AI Agents Course?
A free, self-paced course that takes you from what an agent is to a working agent scored against a public benchmark. It is unusual in teaching three competing frameworks side by side rather than selling one, and it ends in automated evaluation with a student leaderboard plus two free certificates. Prerequisites are basic Python and a rough idea of what an LLM does; Unit 1 includes a recap.
Is Hugging Face AI Agents Course free?
Hugging Face AI Agents Course is free to access.
What level is Hugging Face AI Agents Course for?
Hugging Face AI Agents Course is aimed at a beginner audience. Recommended background: Basic Python — functions, imports, and the ability to read a traceback, A rough working understanding of what an LLM does; Unit 1 includes a recap, A free Hugging Face account and an access token.
How long does Hugging Face AI Agents Course take?
Expect roughly 4 chapters at ~1 week each, 3-4 hours per week (~15-20 hours total), self-paced with no deadline. Most learners work through it at their own pace.
What will I learn from Hugging Face AI Agents Course?
You'll learn: Explain the Thought-Action-Observation loop and identify where each step tends to fail; Write tools as ordinary Python functions and expose them to an agent correctly; Read chat templates and special tokens well enough to debug a broken agent prompt; Build the same agent three ways in smolagents, LlamaIndex and LangGraph and compare the trade-offs; Assemble an agentic RAG pipeline that reasons over retrieval rather than just retrieving; Instrument an agent for observability and evaluate its behaviour, not only its final answer; Fine-tune an LLM specifically for function-calling in the bonus unit; Submit a finished agent for automated benchmark evaluation and see it scored on a public leaderboard.
Topics
Sources
This page was written from 2 sources, 1 on domains other than huggingface.co.