After five years of development, huggingface_hub has reached version 1.0, marking a major milestone in its evolution as the foundational Python library for open machine learning. The release powers over 200,000 dependent repositories and enables access to more than 2 million public models, 500,000 public datasets, and 1 million public Spaces. With nearly 300 contributors and millions of users worldwide, the library has become central to the AI ecosystem. This release introduces breaking changes aimed at supporting the next decade of innovation in machine learning. Key upgrades include migrating to httpx as the core HTTP backend, which brings native HTTP/2 support, true thread safety, and unified synchronous and asynchronous APIs. The legacy hf_transfer library has been fully replaced by hf_xet, now the default for file transfers. Xet operates at the chunk level (64KB), enabling efficient uploads and downloads by transferring only changed portions of large files—a critical improvement for handling massive models and datasets. The command-line interface has been completely redesigned using Typer, replacing the deprecated huggingface-cli. The new hf command features a modern resource-action syntax, sandboxed installers for easy setup across platforms, and built-in autocompletion. It now supports advanced workflows including Spaces deployment, inference endpoints, job management, and full integration with the Hub’s social features like collections, likes, and follow systems. Significant architectural improvements include the removal of the Git-based Repository class and legacy token management patterns. Authentication is now handled through explicit login(), logout(), and get_token() functions. The old InferenceApi has been replaced by the more robust InferenceClient, which integrates seamlessly with multiple inference providers such as Together AI, Replicate, Groq, and SambaNova. Version 1.0 also introduces foundational support for AI agents via Model Context Protocol (MCP) and tiny-agents. Developers can now build and run AI agents using just around 70 lines of Python, leveraging Gradio Spaces as tools and connecting to local or remote MCP servers. Despite the breaking changes, backward compatibility has been prioritized. Most machine learning libraries will work seamlessly with both v0.x and v1.x. The main exception is the transformers library, which requires huggingface_hub v0.x in its v4 series and v1.x in its upcoming v5 release. A detailed migration guide is available to help users transition smoothly. The journey began in 2020 with a simple vision: make sharing machine learning models as easy as sharing code on GitHub. What started as a Git wrapper for the transformers library has grown into a full-featured platform for managing models, datasets, and AI applications. The shift from Git to HTTP in 2022 was a turning point, enabling faster, more reliable interactions without requiring Git LFS. The introduction of Xet in 2024 revolutionized large file handling, with a full migration of over 6 million repositories and 77 petabytes of data completed transparently. Today, huggingface_hub is used by major frameworks including Keras, LangChain, PaddleOCR, YOLO, Google Generative AI, NVIDIA NeMo, and many others. Its impact is felt across research, development, and production environments. The team thanks the global community of contributors and users for their trust and collaboration. While v1.0 marks a significant achievement, it is also a commitment to the future. The library will now focus exclusively on v1.0 and beyond, with older versions receiving only security updates. The goal remains clear: to continue building the open infrastructure that empowers the next generation of AI innovation.