we are using automation API and have a self service no code platform called Q-Cloud. ya the api uses a bit of memory for the pulumi container we have BUT this is only during execution as you mention. we have optimized a few things and works fine for 25 users / occasionally slows down during 5 or so stack updates concurrently. we thing the mem usage may be due to our proprietary layer on top of pulumi api to store each stack no code canvas state locally in a file system. Our stack is nodejs for api but use TS for the program... is your deployment containerized ?