Building an AI agent doesn’t have to mean wrestling with GPUs or spending days setting up a local environment. In a new step-by-step “cookbook,” Hugging Face Press Lead Florent Daudens walks readers through creating a fully functional news research agent using OpenAI’s GPT-OSS open-weight models.
Image source: Envato
The agent can scan headlines, run targeted searches on specific sites, fetch full articles when more depth is needed, and compile concise summaries with clickable source links. All of this happens inside a lightweight Gradio chat interface that runs entirely in the cloud.
Image source: Envato
Daudens positions the project as a ready-made starting point: “If you’ve been wanting to try agents but weren’t sure where to start, this is a complete example you can fork, run, and adapt.”
The guide covers everything from connecting the GPT-OSS 20B and 120B models via Hugging Face’s inference API to defining the agent’s toolset and output format. The code is open, and the design leaves plenty of room for customization.
The full documentation and code are available via Hugging Face Spaces:
https://huggingface.co/blog/
Leave A Comment