Build AI Agents with smolagents
Understand agent architecture, build with smolagents, connect MCP, observe with LangFuse, and share on Hugging Face.
Most "AI apps" are still one-shot chat completions. Useful agents are different: they plan, call tools, run code, delegate to specialists, and loop until the job is done. smolagents is Hugging Face's lightweight framework for building that loop without wiring your own ReAct scaffold.
This is a paid course ($29.99). Lesson 1 is a free preview — buy once for lifetime access to all lessons, or get it included with Gold membership.
Module 1 — Agent Concepts (theory): LLMs, agent definition, workflows vs agents.
Module 2 — Build with smolagents (hands-on): starts at Intro to smolagents — repo labs through MCP, LangFuse, and Hugging Face.
Repo, slides, and setup live in lesson 1.
What you'll learn#
- What makes an agent different from a chatbot — and when the complexity is worth it
- CodeAgent vs ToolCallingAgent, custom tools, and multi-agent orchestration
- Connecting MCP servers you built in MCP Foundations
- Tracing agent runs with OpenTelemetry and LangFuse
- Sharing demos on Hugging Face Spaces
What you'll build#
Not slides. Working agents:
- A minimal CodeAgent solving a multi-step task (
first-agent.py) - A weather analyst with custom tools and Gradio UI (
agent-with-ui.py) - A multi-agent market research pipeline (
multi-agents.py) - An MCP-connected research agent (
agent-with-mcp.py) - An observed, traceable agent run in LangFuse (
agent-with-monitor.py)