smolagents

A barebones open-source library for agents that think in code

Hand-reviewed by the Toolglade team·Updated Aug 2026

agent-frameworks#ai-agents#open-source#code-agents#hugging-face

Free plan Claimed API Self-hosted

About smolagents

smolagents is Hugging Face's minimalist open-source agent library whose agents write actions as Python code, is model-agnostic, and integrates with the Hugging Face Hub for sharing tools.

smolagents is Hugging Face's answer to heavyweight agent frameworks: a deliberately small library, a few thousand lines at its core, that makes it easy to build and run agents in just a handful of lines of code. Its guiding principle is less is more, keeping the abstraction surface tiny while still delivering comparable capability. Its signature feature is native support for Code Agents, where the agent's actions are expressed directly as Python code instead of text or JSON instructions. This code-driven approach tends to be more expressive and composable for tool use. smolagents is model-agnostic, supporting local transformers or Ollama models, many providers on the Hugging Face Hub, and OpenAI, Anthropic, and others via LiteLLM. The library integrates with the Hugging Face Hub so developers can push and pull tools and agents for instant sharing, and it gained thousands of GitHub stars within days of release. It is a strong fit for developers who want a lightweight, transparent, hackable agent foundation rather than a large opinionated framework.

TL;DR

smolagents is Hugging Face's tiny, open-source, code-first agent library that lets agents write Python actions, is model-agnostic, and shares tools via the Hub.

Company overview

smolagents is developed and maintained by Hugging Face, the widely used open-source AI platform. It reflects Hugging Face's strategy of lowering the barrier to agent development with a minimalist, hackable library.

The project was open-sourced and quickly gained thousands of GitHub stars, benefiting from tight integration with the broader Hugging Face ecosystem including the Hub, transformers, and inference providers.

Product features

The library's core is only a few thousand lines and centers on Code Agents that express actions as executable Python. It is model-agnostic, working with local transformers and Ollama models, Hub providers, and external APIs via LiteLLM.

Hub integration lets developers push and pull tools and agents for instant sharing, and the small surface area makes the framework easy to read, extend, and debug compared with larger frameworks.

Target market

smolagents targets developers, researchers, and the Hugging Face community who want a lightweight, transparent foundation for building agents, especially code-driven ones.

Buyer personas

End users

Developers and researchers building agents in Python.

Buyers

Individual developers and small teams (the library is free).

Key influencers

Hugging Face community and open-source contributors.

Ideal customer profile

Developers and research teams already in the Hugging Face ecosystem who want a minimal, code-first, model-agnostic agent framework they can fully control.