rApp

An rApp is the non-RT RIC's equivalent of an xApp — a modular application, but aimed at the jobs that need time and a broad view rather than speed. Training an ML model, managing policy, running analytics across the whole network, recommending configuration tweaks: anything where reacting in seconds is fine and the value comes from seeing the bigger picture. Vendors and operators build rApps against the SMO's framework so this logic, too, stops being locked inside proprietary OSS tooling.

The natural pattern is rApps feeding xApps. An rApp might train a traffic-prediction model overnight, then push it (and a governing policy) down to the near-RT RIC over A1, where an xApp uses it to make fast decisions. So you get a clean division: long-loop intelligence and learning up in the rApp layer, short-loop enforcement down in the xApp layer. If you're coming from the core side, the rApp ecosystem is conceptually close to what NWDAF does for analytics — just scoped to the RAN and the O-RAN framework.