A new open-source framework called smolagents simplifies the creation of AI agents that can write and execute code to perform tasks. Developed by the Hugging Face team, smolagents lets developers define agents that use Python or other programming languages to interact with tools and data, rather than relying on natural language reasoning loops.
The agents can be set up with just a few lines of code, making it easy to prototype and deploy small, efficient assistants. The framework supports multiple backends (OpenAI, Anthropic, Hugging Face models) and integrates with existing codebases via simple tool definitions.
"Smolagents is designed for developers who want their agents to 'think' in code, which can be more precise and faster than generating natural language," the team explained in a blog post.
The project is available on GitHub under an Apache 2.0 license.