Smolagents
Minimal agent framework where agents write and execute their own Python
View on GitHubWhat it solves
The complexity trap in agent frameworks: Smolagents stays small and auditable so you can reason about exactly what your agent is doing instead of debugging a black box.
Overview
Hugging Face's answer to framework bloat: a ~1,000-line agent library where agents solve tasks by writing and running Python code directly rather than calling a fixed set of tools. The whole framework fits in one file you can read in an afternoon. Launched January 2025 and adopted quickly inside HuggingFace's own workflows.
Key facts
- Language
- Python
- License
- Apache 2.0
- Maturity
- Stable
- Maintainer
- huggingface
- Reviewed
- 2026-06-11
Where to start
The guided tour shows how a CodeAgent and a ToolCallingAgent differ in about ten minutes, read this first before choosing an agent type.
docs/source/en/guided_tour.mdTechnologies
More in this domain
LangGraph
Stateful multi-agent orchestration for LLM applications
The state management problem in multi-agent systems: how you build agents that pass context across turns without losing the execution thread.
AutoGen
Multi-agent conversation framework from Microsoft Research
How to coordinate multiple specialized agents toward a shared goal without writing low-level orchestration logic by hand.
CrewAI
Role-based multi-agent orchestration for collaborative task execution
The boilerplate problem in multi-agent setup: defining roles, delegation rules, and task routing without writing a custom orchestration graph.