Strongly seconding the local-hardware route, having done exactly this after the same thing happened to me.
The part that surprised me was how much better the result was. The cloud integration had a delay of a second or two, failed whenever my connection did, and reported state that was sometimes wrong. The local version is instant, works during an outage, and the state comes from a sensor on the actual mechanism rather than from what a server believes.
Worth saying plainly: you are usually not downgrading. You are removing a middleman that was adding latency and a failure mode.
Budget an evening, not a weekend. The wiring is simple; the time goes into mounting things tidily and testing the safety behaviour properly.