Understanding DiskANN, a foundation of the Copilot Runtime

DiskANN (Disk Accelerated Nearest Neighbors) is a key component of Microsoft's Copilot Runtime, providing a new way to manage vector indexes using a mix of in-memory and disk storage. Designed for applications that handle frequently changing data, DiskANN performs searches across vector indexes built from language model embeddings. It supports rapid changes and high search accuracy with lower CPU usage. DiskANN can be implemented using the diskannpy Python library, and Microsoft is expected to expand its usability through additional tools and APIs in the future. The technology is also being employed in services like Cosmos DB and Microsoft 365, ensuring efficient and secure vector searches.