Skip to content

AI9 min read

Siri AI Developer Preview: what App Intents developers need to know

On November 3, 2026 Apple opened a Developer Preview of Siri AI with the 27.3 betas. What was announced, what is still unknown, and how a small studio can prepare its App Intents now.

On November 3, 2026 Apple published Siri AI Is Now Open to All Developers. Starting with the 27.3 betas, any developer can try integrating an app with a new version of Siri. The same day Apple released the 27.3 betas of all its operating systems, and put the technical details into a new documentation article, Siri AI Developer Preview.

We ship our own iOS apps (calls, VPN, utilities), so we read the announcement and the documentation pages behind it. This post sums up what Apple says, where the limits are and what a small studio can do now. Facts come from Apple's pages; our interpretation is marked as such. We have not built a Siri AI integration yet, and nothing below is a result of our own testing.

What Apple announced

The Developer Preview comes with the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas. Apple describes Siri AI as a new version of Siri that is "profoundly more capable, conversational, and personal" (Apple's announcement). It is meant to be integrated into the operating systems and into apps, and Apple says thousands of developers have already built Siri integrations for things like directions, to-do lists, messages and audiobooks.

The key points for developers:

  • Apps bring content and actions to Siri AI through App Intents, the framework many of us already use for Shortcuts and widgets.
  • Siri AI can use personal context, onscreen awareness and natural language understanding, so people can ask for custom actions in their own words: check a delivery, book a table, place an order.
  • Developers can teach Siri AI skills, which combine several actions into one experience for a complex request.
  • Siri AI can surface any entity from the Spotlight semantic index, not only entities that fit a predefined schema.
  • Apple's own apps are getting the same capabilities: Calendar, Files, Finder, Journal, Mail, Maps, Messages, Music, Notes, Photos, Preview, Reminders, Safari, Shortcuts and Weather.

The documentation adds an important detail: you do not need a new entitlement or a separate developer program to take part in the preview. It is the existing App Intents code, extended.

How an app plugs in

The documentation article on bringing custom actions and content to Siri is the practical core. Its main rule is that Siri can only find content and perform actions that the app makes explicitly available. In short:

  1. Tools. To expose a custom action, add the @AppIntent(.tool) macro to a new or existing app intent. It makes the intent a tool Siri can call. Intents that already conform to a schema in a supported primary domain do not need it, and an intent with no real perform() work, such as a plain widget configuration intent, should not get it.
  2. Titles and descriptions. Siri tells intents apart by their title and description, and fills parameters from the title and description of each parameter. Entities need the same: add a description to every @Property, @DeferredProperty and @ComputedProperty.
  3. Finding content. Make app entities available to Spotlight so Siri can find them. As a fallback, the system can also use an EnumerableEntityQuery that you provide.
  4. Context and cues. Associate views and user activities with entities for onscreen context ("this photo"), and donate intents when people perform actions so Siri has behavioral cues.
  5. Moving data between apps. Entities that travel from one app to another need Transferable representations.

Here is the documentation's own example of a tool intent for a shopping app:

swift
@AppIntent(.tool)
struct AddToCart: AppIntent {
    static let title: LocalizedStringResource = "Add to Cart"
    static let description = IntentDescription(
        LocalizedStringResource("Adds a product to the shopping cart.")
    )

    @Parameter(title: "Item", description: LocalizedStringResource("The product to add"))
    var item: ProductEntity

    @Parameter(title: "Quantity", description: LocalizedStringResource("How many to add"), default: 1)
    var quantity: Int

    func perform() async throws -> some IntentResult {
        try await Cart.shared.add(item, quantity: quantity)
        return .result()
    }
}

For users, this means they do not have to name your app while it is in the foreground, while a Live Activity is running or when a conversation already mentioned it. In the background, naming the app is enough. Siri can also hand content from one app to another and show results in card snippets built from your own views.

Privacy is part of the API

By default, app entities do not require authentication before Siri or Shortcuts present their data. For sensitive properties, the app declares an IntentAuthenticationPolicy, for the whole entity or for a single property. Apple's example is a bank account whose balance requires local device authentication. A property can also be marked with visibility: .internal, so only the app and its extensions can read it. One detail: a property-level policy applies to responses Siri provides, but if an intent returns sensitive data as its result, the policy must be set on the intent itself.

Skills: guidance, not scripts

A tool intent does one thing. When a request needs several, app skills tell Siri how the pieces fit together. A skill is a Markdown file named SKILL.md with two frontmatter fields, name and description, followed by free-form instructions of up to 10,000 characters. Skills live in a Skills.skillsbundle that Xcode adds to the app target, one folder per skill, and an app target can contain only one such bundle.

Apple is explicit that skills are not deterministic. Siri reads the description first to decide whether a skill matches, reads the body only after that, and does not run the listed tool intents in order. A skill is a hint about which intent to prefer, what to avoid and what domain knowledge applies, not a workflow engine.

Timeline and status

  • The preview is available now, in the 27.3 betas. Apple calls it an early look.
  • Apple says it will ship to customers "in a few months".
  • App Store submission for Siri AI experiences opens in the spring.
  • Apple says it will share more details about terms and usage policies for experiences enabled through Siri AI. It has not published them yet.
  • Apple Intelligence Group Labs for questions run on November 4 (9 to 10 a.m. PT) and November 5 (6 to 7 p.m. PT).

The testing rules are narrow. You can build with Xcode, sign with a development profile and run on a supported device, or distribute to internal testers in TestFlight. The preview is not available to external TestFlight testers or in App Store builds, and Apple warns that App Intents API and Siri behavior may change during the preview. For skills, the documentation says to test on a real device, not in the Simulator.

Limits and conditions

The footnotes in the announcement are worth reading before planning anything:

  • Languages. Available now: English, French, Japanese, Korean, Portuguese and Spanish. Coming soon: Traditional Chinese, Norwegian, Turkish and Vietnamese. Neither Serbian nor Russian is on either list.
  • EU. Siri AI is currently unavailable in the EU on iOS, iPadOS and watchOS. Apple says it is working with the EU to find a path forward.
  • Age. Not available to users under 13.
  • Devices. Requires an Apple Intelligence-enabled device set to a supported language. The announcement does not list models.
  • Usage limits. Some server-side Apple Intelligence features have daily limits, including Siri AI. Limits depend on the feature, request complexity, system demand and policies, and Apple gives no numbers. Expanded access "will be available for a fee in the future", with no price yet.

What this means for a small studio (our interpretation)

If an app wants to be part of Siri AI, the work is App Intents work. That is good news for a small team: there is no separate SDK, and titles, descriptions, entities and Spotlight indexing also help Shortcuts and system search. We would start there, because it pays off even if Siri AI changes.

The first question is which actions are safe to expose. In a calls app, finding things the user already sees is an easy candidate; starting a call by voice is a stronger action that we would only expose with a confirmation. In a VPN app, connecting and disconnecting change the device's network state, so what Siri may do on its own needs a deliberate decision. In a utility app, well-described entities are a cheap first step. None of this is tested; it is our reading of the documentation and our own risk assessment.

Privacy needs its own design pass. Our guide to App Privacy labels covers what an app tells Apple about its data. With Siri AI a second question appears: which data does the app let the system read and speak aloud? Entity properties are open by default, so review every property, mark internal ones as internal and require authentication for anything personal. For private communication apps, like the one in private calls without a phone number, we would start restrictive and open properties one at a time.

Languages are the other practical point. This site is available in Russian and Serbian, and many of our users speak them. Neither language is on Apple's current or announced list, so Siri AI features may not reach a large part of that audience at first.

What we do not know yet

Apple has not published the terms and usage policies for Siri AI experiences, so we cannot say which kinds of actions will be allowed, how App Review will treat them, or whether the App Review Guidelines will get new sections. We also do not know the daily limits, the price of expanded access, the list of supported devices, or whether the EU restriction will change before launch. The API itself may change during the preview.

Our reading: do the preparation that is useful on its own, and wait for the terms before building anything that depends on them.

What to do

  • Read the Siri AI Developer Preview article and the two guides it links to.
  • Inventory your app's intents and entities, and decide which actions Siri may perform and which it must never touch.
  • Add clear title and description values to every intent, parameter and entity property.
  • For sensitive data, set an IntentAuthenticationPolicy or mark properties as internal. Check the intent result too, not only the properties.
  • Index the entities people would ask about in Spotlight.
  • Try the Siri Integration Advisor in Xcode and the Siri Evaluations framework on a real device with a development build.
  • If you write a skill, keep it short, describe intents by what they do, and do not rely on it for strict ordering.
  • Do not announce Siri features to users or in store metadata until Apple publishes the terms.
  • Check the languages and regions you serve against Apple's list, and plan fallbacks for unsupported ones.
  • If you can join the November 4 or 5 labs, bring questions about terms, review and usage limits.

Sources

ShareXLinkedIn

ONLYAPPS Team

Novi Sad, Serbia · onlyapps.org

Contact

We use analytics cookies only with your consent. Essential cookies keep the site working. Privacy policy