I have the same question, and after watching the sessions I think the answer is clearer than the thread suggests, which is part of why I wanted to raise it here.
I watched Build intelligent Siri experiences with App Schemas along with sessions 345 and 343. My read is that they're consistent and explicit: to expose your entities and actions to the new Siri and the agentic features, those entities have to conform to the published App Schemas. If your data doesn't fit one of the whitelisted domains, there's no supported path.
Here's the concrete version. Apple's own sample, Defining app entities for your custom data types, builds a hiking app with a TrailEntity. A TrailEntity maps to none of the published domains. So a user with that app can't ask Siri to compare routes or pick a trail, and there's nothing the developer can do to enable it. Most apps with interesting custom data are in the same position. As it stands, that leaves a large part of the third-party ecosystem out of the agentic experience, and it leaves users believing Siri is still the unreliable assistant it has long been reputed to be.
I want to be constructive about this. The beta only just shipped, so this is still a decision that can change before the public release.
A few things that would make a real difference:
- Give developers clear guidance on how to annotate entities and actions now, even without a matching domain, so the work becomes useful when the restriction is lifted.
- Let developers define or propose their own domains.
- Let the on-device model infer a domain from a well-annotated entity when there's no exact match.
I filed this as FB23018652 with the TrailEntity example. If this affects your app too, filing your own feedback and boosting this thread would help it get a real answer while there's still time to change it.
