Small agents. Big room to experiment.
A starting point for tool-using agents, with a little less ceremony and a lot more intentional supervision.
START.
BIG IDEAS.Ideas are better when shared.
Meet smolagents
Hugging Face smolagents is a library for building agents, including agents that express actions in code. Explore the official repository and documentation before choosing an execution setup.
Give the agent a narrow job
Start with a low-stakes task such as summarizing a small set of public documents. Grant only the tools it needs. Use isolated execution and review outputs before allowing external actions.
Share the whole experiment
Tell the community what the agent was asked to do, which tools and model it used, where it failed, and how a human checked the result. A thoughtful failure report counts as a valuable contribution.
Go to the source
Keep the idea moving.
Have a question, improvement, or your own experiment? Bring it to the repository.
Discuss on GitHubSuggest an edit