“In distinction, Azure Database and Cosmos DB for PostgreSQL stays the general-purpose managed Postgres for conventional apps and the distributed, sharded possibility for horizontally scaled multi-tenant workloads respectively,” she stated.
“HorizonDB is the AI-era Postgres tier, the place transactional modernization and vector-driven purposes lastly land in the identical engine. It’s meant to be the place the place you each modernize legacy transactional techniques and bolt on RAG, brokers, and embeddings-heavy options with out stitching collectively separate databases and ML endpoints,” stated Walter.
One of many issues that has made PostgreSQL so standard with AI builders is its assist for vector indexes, stated Microsoft, including that in Azure HorizonDB it’s extending that assist with superior filtering capabilities within the DiskANN vector index that “allow question predicate pushdowns straight into the vector similarity search”. This frequent optimization method filters information as early as potential within the pipeline to keep away from pointless information processing and transmission.