Making Your App Usable by Siri, Gemini and AI Assistants: App Intents and AppFunctions
How Siri, Gemini and AI assistants act inside apps through App Intents and AppFunctions, which actions to expose and how to keep them safe.
Quick answer
AI assistants are moving from answering questions to doing things inside apps. On iOS, App Intents describe your app's actions and content so Siri, Spotlight, Shortcuts and Apple Intelligence can use them; iOS 27 adds entity and intent schemas that let Siri act on your content without predefined phrases. On Android, AppFunctions let apps expose functions that Gemini and other agents can discover and call, effectively turning the app into an on-device MCP server (experimental preview in 2026). Start by exposing a few high-value, low-risk actions and your key content, require confirmation for anything consequential, and design the actions to follow the same rules as your UI.
Why this matters now
On the web, AI agents are starting to browse sites and complete tasks for people (see how AI agents use websites). On phones the same shift is happening through the operating system. Rather than a user opening five apps, they ask the assistant to "reorder last week's groceries", "show my next booking" or "send the invoice to the client", and the assistant calls the app that can do it.
Apps that describe their actions to the system become the ones assistants use. Apps that do not are reachable only by a person tapping through screens, which is fine today but increasingly a disadvantage.
Apple: App Intents and Siri
App Intents is Apple's framework for exposing actions (intents), content (entities) and queries to the system. It already powers Shortcuts, Spotlight actions, widgets, Control Center controls and Siri. Apple's WWDC26 guidance for iOS 27 extends it for Apple Intelligence:
- Entity schemas contribute your app's content to the Spotlight semantic index, so Siri can surface it with attribution back to your app
- Intent schemas let people act on that content in natural language, with no specific phrases to define and no code changes as Siri's language understanding expands
- View Annotations map on-screen views to entities, so people can refer to and act on what is on screen conversationally
- App Intents Testing framework validates the integration through real system pathways without UI automation
Android: AppFunctions and Gemini
AppFunctions is an Android platform API with a Jetpack library. An app declares self-describing functions, which register with an OS-level registry so that authorized callers (assistants such as Gemini, and other agents) can discover and execute them, in the background and without opening the UI. Google describes it as allowing your application to act as an on-device Model Context Protocol (MCP) server.
Status matters here. At Google I/O 2026, AppFunctions was in experimental preview, with a test agent for debugging and an early access programme for production deployment; it targets Android 16 and later. Samsung's Gallery integration with Gemini on the Galaxy S26 series was an early public example. Plan for it, prototype it, but expect APIs to change before general availability.
Worth noting
Platform status reflects Apple's WWDC26 iOS guide and the Android Developers blog from Google I/O 2026, as of October 2026. Check current documentation before scheduling production work.
Which actions to expose first
Do not try to expose everything. Pick actions users repeat often, that are tedious to reach by tapping, and that are safe or easily confirmed.
| App type | Good first actions | Hold back or require confirmation |
|---|---|---|
| Ecommerce | Track order, reorder a past order to cart, check stock | Placing paid orders, changing addresses |
| Travel and hospitality | Show next booking, check-in details, add to calendar | Cancellations with fees, payments |
| Banking and fintech | Show balance summary, recent transactions, spending category | Transfers, payee changes (with strong authentication) |
| Health and fitness | Log water or workout, show today's plan | Anything clinical; sharing data |
| Productivity and B2B | Create a task, find a document, show today's schedule | Deleting data, sending external messages |
| Food and delivery | Reorder favourite, track delivery | Payment changes |
Designing actions an assistant can use
Assistant-callable actions are tools for an AI, so the principles of good tool design apply: clear names and descriptions, typed parameters, results that are easy to present, and errors that explain what to do next. Our guide to AI agent tool design covers these in depth.
- Name actions by user intent ("Reorder groceries"), not internal function names
- Use typed parameters and entities (an order, a booking) rather than free text
- Return a result the assistant can show or speak in one or two sentences
- Enforce the same authentication, permissions and business rules as the UI
- Make write actions idempotent so a repeated request does not duplicate an order
- Provide a deep link to the relevant screen for anything that needs the full UI (see mobile app deep linking)
Privacy, security and confirmation
Exposing actions to the system means content and capabilities can be reached outside your UI. Decide deliberately what each action reveals: an assistant reading out an account balance on a locked phone is a different risk from showing it in-app after biometrics. Respect the platform's authentication requirements, require confirmation for purchases, transfers, cancellations and outbound messages, and log assistant-initiated actions separately so you can audit and analyse them.
On Android, AppFunctions require callers to hold a designated permission; on iOS, the system mediates access. Neither removes your responsibility to apply your own business rules inside the action.
Want your app ready for AI assistants?
ZSpace Labs adds App Intents, AppFunctions, deep links and the backend changes behind them to iOS, Android and React Native apps. See mobile app development.
Measuring assistant usage
Track how often each action is invoked through the system versus the app UI, completion and error rates, and what users do next. These numbers show which actions deserve investment and whether assistant users retain differently. See mobile app analytics for event design.
Cross-platform apps
App Intents are implemented in Swift and AppFunctions in Kotlin, so React Native and Flutter apps add a thin native layer for each platform. Keep the business logic shared (often on your server or in a shared module) and make the native layer a small adapter. The two platforms' models differ: Apple organizes around schemas, entities and the semantic index; Android around discoverable functions. Design your action catalogue once, then map it to each.
A practical roadmap
- Now: list the 5–10 actions users repeat most; check which are already exposed as App Intents, shortcuts or deep links
- Next release: add App Intents and entities for the safest high-value actions; add deep links for everything else
- Prototype AppFunctions for the same actions on Android test devices
- Add confirmation and logging for any action that changes data
- Measure usage and errors, then expand the catalogue
Conclusion
Assistants are becoming another way people use apps, and both Apple and Google are building the plumbing into their operating systems. Apps that expose a small set of well-designed, safe actions and their key content will be the ones assistants can use. Start with App Intents on iOS today, prototype AppFunctions on Android, keep consequential actions behind confirmation and treat the action catalogue as part of your product. For the on-device side of mobile AI, read on-device AI in mobile apps.
Common questions.
Apple's framework for describing your app's actions and content to the system, so they can be used by Siri, Shortcuts, Spotlight, widgets and Apple Intelligence. In iOS 27, entity and intent schemas let Siri surface your content and act on it without fixed phrases.