postgresml

Why we included this project

PostgresML keeps machine learning inside PostgreSQL instead of making you stand up a separate model-serving service. It installs as an extension and gives you SQL functions for training and running classification and regression models, pulling embeddings from Hugging Face transformers, and completing the chunk-embed-rank-transform steps of a RAG pipeline in a single query. Because inference runs where the data already lives, you avoid the latency and overhead of shipping rows over HTTP to an external endpoint, and sensitive records never have to leave the database. That makes it a practical pick for teams that want vector search or LLM features bolted onto their existing operational database rather than introducing another data platform. If you like prototyping in plain psql, GPU acceleration is available on the infrastructure you already run.