To be a little more clear, KumoMTA is an On-Premises Message Transfer Agent designed to be run efficiently in a cloud that can be public (AWS, Azure, GCP, etc.) or private (vSphere, Kubernetes, Docker) or just downloaded and installed on your local physical server. The On-Premesis part means you own and run it instead of paying a SaaS provider for an email relay service. This is similar in the primary function to several other on-premises MTAs in that it will pass email from one place to another, but KumoMTA is so much more.
For the uninitiated, MTAs are the infrastructure that passes email from place to place on the Internet. In the world of MTAs, there are on-premises (you download software and run it) and SaaS (Software as a Service) options. KumoMTA is the former solution where you can install the software onto a server you manage and run it any way you like. This means you do not have to pay a monthly fee to a SaaS company to manage your email delivery.
For that matter, you have no license fees at all.
KumoMTA is an OpenSource project that a qualified person can download and install on almost any Linux OS without ever speaking to a salesperson or paying any license fee. While this may sound similar to Exim or Postfix, you would be hard-pressed to get either of those to process multiple millions of messages per hour and handle multiple domains and IPs out of a single node.
And that brings me to awesomeness point three - the need for speed. We will publish a focused performance blog at a future date, but as a teaser, I recently ran a comparison between PostFix and KumoMTA using identical environments. In both cases, I generated 100,000 100KB messages injected to the localhost for processing and then forwarded to a third-party smarthost that would then blackhole the message (throw it away). That configuration measures the raw processing power of the MTA without any feedback or bounce processing. While I could get the Postfix server up to 500,000 messages/hour (100,000 in 12 minutes), KumoMTA was able to handle the same workload in just over one minute. Both tests were done on the same Azure VM template with 8vCPU, 16Gb RAM, and 30Gb gp2 storage. This test shows that a very reasonably sized cloud instance can perform well into the 4MMH and higher performance range in a real-world scenario. If it helps, here is a graphic: