Designing the hf CLI as an agent-optimized way to work with the Hub

The hf CLI, the official command-line interface for the Hugging Face Hub, has been re-engineered to cater to both human users and increasingly, AI coding agents. This strategic redesign addresses the distinct preferences of each user type, ensuring optimal interaction with the Hub.

Before this redesign, the hf CLI primarily served human users. However, with the rising adoption of AI agents like Claude Code and Codex, a need emerged for an interface that could also efficiently support automated operations. The new hf CLI minimizes token usage for agents on complex, multi-step tasks, demonstrating up to a 6x reduction compared to baseline methods.

Key to this dual functionality is the hf CLI's ability to detect when an AI agent is in control. It identifies agents through specific environment variables and then adapts its output accordingly. For humans, the CLI provides rich, color-coded terminal output with clear hints. For agents, it delivers concise, structured data without visual clutter, ideal for programmatic parsing.

This agent-aware output mechanism was introduced in hf v1.9.0 and has been progressively rolled out. The system automatically adjusts the output format, though users can manually specify their preferred format. This approach ensures that both human developers and AI agents receive information in the most suitable and efficient manner for their specific workflow.

Agent traffic on the Hugging Face Hub has shown significant growth, with Claude Code and Codex leading in distinct users and request volumes. This trend underscores the importance of an agent-optimized CLI, as AI agents are becoming a standard method for interacting with the Hub, contributing to a substantial increase in overall activity.