Siri AI Gains New App Actions and Languages in iOS 27.3

Overview of the iOS 27.3 Siri AI Expansion

Apple’s latest incremental release, iOS 27.3, marks a decisive step toward making Siri a true productivity partner rather than a simple voice‑activated assistant. The update opens Siri’s large‑language‑model (LLM) engine to a broader set of system apps, introduces four new language locales, and provides a Developer Preview that lets any third‑party app expose custom actions through App Intents. In practice, users can now ask Siri to perform multi‑step workflows that previously required manual navigation—think “unsubscribe me from newsletters in Mail” or “clear duplicate tabs in Safari.” The move signals Apple’s confidence that on‑device AI can handle more sophisticated, context‑aware tasks while preserving privacy.

Technical Deep Dive: App Intents and On‑Device LLM

App Intents Architecture

Apple’s App Intents framework, first hinted at in earlier iOS releases, is now the primary conduit for developers to surface app functionality to Siri AI. An intent is a declarative description of an action, its parameters, and the expected result. Under the hood, the system compiles intents into a SiriKit Extension that runs in a sandboxed process, allowing the on‑device LLM to invoke the intent without leaving the device.

Key technical points:

• Intent Definition Files (IDL): Written in Swift, these files describe the intent’s name, input types (e.g., Date, Location), and output schema.
• Dynamic Parameter Resolution: Siri can ask follow‑up questions to disambiguate inputs, leveraging the LLM’s conversational abilities.
• Privacy‑First Execution: All intent handling occurs locally; no raw user data is transmitted to Apple’s servers unless the developer explicitly opts into cloud processing.

Expanded Language Support

iOS 27.3 adds Traditional Chinese, Vietnamese, Turkish, and Norwegian to Siri’s supported locales. The on‑device model was retrained with multilingual corpora, ensuring that the same contextual reasoning applies across languages. For developers, this means that an intent defined once can be automatically localized, provided the app supplies appropriate LocalizedStringKey resources.

System‑App Enhancements

Apple has instrumented 15 core apps with new Siri‑driven actions. Below is a non‑exhaustive list of the most impactful capabilities:

• Calendar: “Schedule a meeting with Alice next Thursday at 3 pm” now creates an event, invites participants, and adds a location based on prior patterns.
• Mail: Unsubscribe from newsletters, undo a sent email, and draft replies using natural language prompts.
• Safari: Clear browsing history, manage tabs, save pages to Reading List, and close duplicate tabs.
• Files: Organize files into folders, delete items not accessed

not accessed for a configurable period, and even suggest archiving them to iCloud Drive.

• Photos: “Create a slideshow of my beach photos from last summer” now pulls the relevant images, applies a transition style, and queues the slideshow in the Photos app.
• Music: Users can ask Siri to “Generate a workout playlist that mixes my favorite EDM tracks with new releases from this week,” and the AI will curate a dynamic playlist, add it to the Library, and start playback.
• Notes: “Summarize the meeting notes I took on Tuesday” triggers an on‑device summarization model that extracts key points and appends a concise summary to the original note.
• Reminders: “Remind me to water the plants every three days starting tomorrow” creates a recurring reminder with a custom icon and location‑based trigger.
• Weather: Beyond adding locations, Siri can now “Turn off severe‑weather alerts for my weekend trip to Denver” or “Set a daily forecast notification for 7 am.”
• Maps: “Find a coffee shop with outdoor seating within a 5‑mile radius and add it to my Favorites” combines POI search, attribute filtering, and list management in a single request.
• Messages: “Send a quick thank‑you note to Maya with a GIF of a high‑five” lets Siri pick an appropriate GIF from the built‑in library, attach it, and send the message.
• Journal: “Log a gratitude entry about today’s project milestone” creates a new journal entry with suggested tags and a mood emoji based on the user’s tone.
• Preview: “Convert the PDF I’m viewing into a searchable document and save it to Files” runs OCR on the document, embeds the text layer, and stores the result.
• Shortcuts: Siri can now invoke any Shortcut by name and automatically fill in its parameters, effectively turning every Shortcut into a conversational action.

Developer Preview: Bringing Your App Into Siri AI

Apple’s Developer Preview for iOS 27.3 is available through the Apple Developer portal. It provides:

1. App Intents SDK – A Swift‑first API that lets you declare intents, expose rich parameter types (including custom structs, dates, locations, and even media assets), and define fallback UI for cases where the LLM cannot resolve a request.
2. Siri AI Skill Builder – A low‑code interface where developers can compose multi‑intent “skills” that chain together actions. For example, a travel app could expose a skill that books a flight, reserves a hotel, and adds the itinerary to Calendar—all triggered by a single user utterance.
3. Testing Harness – An on‑device simulator that lets you type or speak sample queries and see how the LLM maps them to your intents, complete with confidence scores and disambiguation prompts.
4. Analytics Dashboard – Aggregated, privacy‑preserving metrics (e.g., intent usage frequency, success rate) to help developers iterate on their AI experiences without exposing raw user data.

Quick Start Example

import AppIntents

struct AddExpenseIntent: AppIntent {
 static var title: LocalizedStringResource = "Add expense"
 static var description = IntentDescription("Record a new expense in your finance app.")

 @Parameter(title: "Amount")
 var amount: Double

 @Parameter(title: "Category", default: .food)
 var category: ExpenseCategory

 @Parameter(title: "Date", default: .now)
 var date: Date

 func perform() async throws -> some IntentResult {
 // Call your app's expense‑recording API
 try await ExpenseManager.shared.add(amount: amount,
 category: category,
 date: date)
 return .result(value: "Expense added")
 }
}

Once registered, users can say, “Hey Siri, add a $45 dinner expense for yesterday,” and the LLM will automatically fill the amount, category, and date parameters, invoking the intent without any additional steps.

Skills: Combining Multiple Intents

Apple introduced Skills as a way to bundle several intents into a cohesive workflow. A skill can include conditional logic, loops, and user‑prompted branching. For instance, a “Plan a road trip” skill might:

1. Ask for the destination city.
2. Retrieve driving directions via the Maps intent. 3.

Read the full breakdown originally published at https://ltdeveloperblogs.github.io/posts/siri-ai-gets-more-capable-in-ios-273/