A model-deployment platform for running custom or open-source models (including LLMs) in production, handling autoscaling, GPU provisioning and inference optimization so a team doesn't manage its own inference infrastructure.