Hugging Face
huggingface_hub
The library shifted toward making the Hub accessible as a complete platform—not just a data source, but an infrastructure layer with persistent storage, compute, and AI-agent integration. Buckets arrived as S3-like object storage for checkpoints and artifacts without version control overhead, while Spaces and Jobs gained volume mounting to access models and datasets as direct filesystem mounts. The CLI underwent a complete agent-aware redesign: it graduated to pip-installable extensions, gained a skills system for AI coding assistants, and expanded discovery commands for models, datasets, and spaces alongside new management tools for jobs hardware and dev mode. Recent patch work in v1.9 solidified the volumes API contract after initial rollout.
Volume mounting for Jobs, completion of the Papers CLI, and CLI enhancements dominated March. Jobs gained the ability to mount Hugging Face repositories and storage buckets directly as container volumes, enabling workflows like querying datasets or persisting checkpoints without explicit downloads. The Papers CLI reached full functionality with search and structured output, while the CLI layer matured significantly—extensions now support pip-installable Python packages with isolated dependencies, hf extension install accelerated through uv integration (6.7x faster), and new command groups for Spaces dev mode, discussions, and webhooks shipped alongside HfFileSystem bucket support.