Hello folks. I'm aware that Netflix has internally developed the capability to create live endpoints from within Metaflow that serve metaflow artifacts. As I understand it, the necessary service infrastructure for that is managed by Metaflow. This also seems to have been on one version of the GSoC ideas list. I see that there are relatively new additions in the metaflow-netflix-extensions (and this open PR github.com/Netflix/metaflow-nflx-extensions/pull/86) that implement many of the mentioned steps. Are there any plans to make the full FastAPI backend infrastructure also available?

We have a couple internal backend implementations that do what you're referring to. The code in metaflow-nflx-extensions is missing a corresponding non-Netflix compute provider to make it work in oss. PR 86 is one attempt to fill that gap using grpc/modal, others can be supported. We are happy to accept PRs if you want to take this up

Thank you @hundreds-rainbow-67050, I realise that I phrased my question very unclearly. I understand that

metaflow-functions

still needs an OSS compute provider, and that PR #86 explores options such as gRPC and Modal. My question is whether Metaflow might also manage the serving infrastructure lifecycle itself: provisioning, updating, scaling, monitoring, and deleting a persistent endpoint. Similar in spirit to what it does for workflows. Is the intended OSS model for Metaflow to manage that deployment, or only to provide the portable function/runtime while users manage the serving platform separately?

The intended model is for metaflow to provide constructs to manage the deployment, in particular seamless integration with flows and binding flow artifacts to the serving platform (we already have this internally at Netflix)

Cool, that would be really great to see!