Given a query (e.g., “who won the first Nobel Prize?”), knowledge-intensive natural language processing (KI-NLP) systems retrieve relevant information from large corpora of information. To advance KI-NLP research, Meta AI has an open-sourced Sphere, a knowledge source containing 134 million documents split into 906 million passages. Unlike existing tools, Sphere uses open web data rather than proprietary search engines so that users may control the corpus when experimenting with new information retrieval techniques and architectures (e.g. dense retrievers). The Meta AI team has also released the indices of Sphere, including a dense index that stores vector representations of the corpus documents. This index is compatible with distributed-faiss, which apportions indices across multiple machines.