Postgresml

PostgresML is a Postgres database with GPUs attached for machine learning and AI applications. It keeps vectors, models and application data in one place, so a team can create embeddings, index, filter and rank vectors, and run inference and grounded answers directly in SQL instead of operating a separate vector database, embedding service and LLM API. It is aimed at engineers building retrieval augmented generation and other ML features who want fewer moving parts and less data crossing vendor boundaries. The home page offers a live demo for text generation, embeddings, summarization and translation using hosted open models such as Llama 3.1, Mixtral, Mistral and several embedding models, with Python, JavaScript and SQL examples. Vector search uses pgvector for KNN and approximate nearest neighbor queries, and the site links to the Korvus SDK as an easier route to RAG. The vendor reports its own comparisons of 4 times faster than a Hugging Face plus Pinecone chatbot and 42 percent lower vector database cost than Pinecone. Published pricing offers $100 in free usage credits on signup. Serverless is pay per use, from $7.50 per query hour, with storage at $0.25 per GB per month and embeddings included with queries. Dedicated starts from $0.60 per instance hour, with committed use discounts, any Hugging Face model and deployment on major clouds. Enterprise is custom priced and adds VPC deployments, multiple GPUs, custom SLAs and premium support.