Siri AI Opens Up to All Apps: Developer Preview with App Intents and Skills

Siri AI can now be controlled from any app, though this is currently only in the testing phase. With the first betas of iOS 27.3, iPadOS 27.3, macOS 27.3, and visionOS 27.3, Apple launched a Developer Preview on November 3, 2026, allowing developers to integrate their own actions and content with the new voice assistant via App Intents. This feature won’t be available to users for a few months; submissions to the App Store will be accepted starting in the spring. And in the EU, Siri AI remains disabled on the iPhone.

Key Points at a Glance

  • Any app can share its own actions and content with Siri AI—no longer just apps that fit into Apple’s predefined schemas.
  • A new feature is “Skills”: Markdown files that developers can use to teach Siri AI multi-step workflows.
  • Testing is limited to internal channels (developer builds and internal TestFlight testers). App Store submissions will begin in the spring, and users will gain access to the features in a few months.
  • In the EU, Siri AI is still not available on iPhone, iPad, and Apple Watch, though it is available on Mac and Vision Pro.
  • German is not yet among the supported languages.

We provided a detailed summary of what Siri AI can do when it was unveiled in June: Apple Unveils Siri AI: The Voice Assistant Is Becoming More Personal, Smarter, and Significantly More Powerful. This update marks the next step: opening up the platform to third-party apps.

What’s Changing with the Siri AI Developer Preview

Until now, Siri AI could only interact with third-party apps via so-called app schemas. These are predefined templates from Apple for typical tasks. According to Apple’s documentation, there are 13 such domains that work with Siri and Apple Intelligence, including Audio, Calendar, Mail, Messages, Notes, Photos, and Video. When Siri AI launched on September 14, for example, users could send a WhatsApp message or play an audiobook on Audible. Tracking a shipment in a delivery service app, however, wasn’t possible because no schema existed for that purpose.

It is precisely this limitation that is now being removed. Using the App Intents framework, any app can pass its own custom actions and content to Siri AI. Apple cites examples such as checking a delivery, making a table reservation, or placing an order. The integration leverages all of Siri AI’s core capabilities: understanding personal context, on-screen awareness (i.e., recognizing content on the display), and natural language. According to Apple, no new entitlement or separate developer program is required.

Tool Intents: A Swift Macro Turns Actions into Tools

Technically, developers mark an intent with the @AppIntent(.tool) macro. This makes it conform to the new ToolIntent protocol and appears to Siri as a tool that the model can invoke on its own. Anyone who has worked with tool calling in large language models will recognize the principle immediately. Intents that already conform to a schema supported by Siri remain unchanged. Intents that previously ran only in Shortcuts (such as a new journal entry) can be upgraded using @AppIntent(.tool, schema:).

It’s interesting to note what Siri looks for when making its selection: the title and description of the intent, as well as each individual parameter. From this, the model deduces what an action does and how it differs from similar actions. Well-formulated metadata thus becomes just as important as the text on an App Store page. Another new feature is a generatable type for entities: For example, Siri can directly assemble a flight search from the sentence “Economy flight from San Francisco to Los Angeles next Friday” without first asking the app for data.

Skills: Markdown Instead of Program Code

For more complex requests, Apple is introducing Skills. A tool intent performs exactly one action. A skill, on the other hand, is a set of instructions in natural language that explains to Siri how multiple tools interact, which one takes precedence in case of conflicts, what should be avoided, and when it makes sense to ask for clarification. Apple’s example is a shopping skill that links search, shopping cart, payment and shipping methods, review, and purchase confirmation.

The format is pleasantly simple. For each app target, there is a Skills bundle with one folder per skill, each containing a SKILL.md file. Two fields are required:

  • name: 1 to 64 characters consisting of lowercase letters, numbers, and hyphens; must match the folder name
  • description: a single line with up to 1,024 characters

The actual text can be up to 10,000 characters long. Siri first reads only the description to determine whether a skill matches the request and loads the rest only afterward. Xcode checks and compiles the bundle during the build. According to the documentation, the entire system is based on a subset of the open Agent Skills format from agentskills.io, though without scripts or reference files.

Important for developers: A skill is a suggestion, not a program. Siri does not rigidly process the listed tools in sequence, and deterministic behavior cannot be explicitly enforced. Ultimately, the model determines what happens in the conversation.

Spotlight, Screen Content, and Sensitive Data

To enable Siri to find content within an app, developers store their entities in the local Spotlight index. Siri AI can now display any entry in the semantic Spotlight index, even if it doesn’t fit into an existing schema. Onscreen Awareness also works with custom entities, so you can refer to what’s currently visible in the app during a conversation. Through so-called “donations,” an app also reports which actions you perform regularly. Apple Intelligence uses this information to identify routines and can, for example, proactively suggest checking the morning weather in a specific app.

Developers retain control over sensitive data. They can require authentication for an entire entity or individual properties, or mark properties as “internal” so that neither Siri nor Shortcuts can access them. Apple’s banking example: Siri is allowed to know that an account exists, but displays the nickname only after authentication and the account balance only after unlocking the device itself. The bank routing number remains entirely within the app.

When You No Longer Need to Name the App

A small but important detail in everyday life: In many cases, you no longer need to say the app’s name. This applies as long as the app is running in the foreground, one of its Live Activities is active, or you’re continuing a conversation in the Siri app where the app was already mentioned. Apple’s example is a ride-hailing app: You order a ride, track it via Live Activity, and then simply tell Siri to change the pickup location. If the app is running only in the background, you’ll still need to say its name.

Cross-app requests are also possible. For example, Siri can find your favorite sneakers in a shopping app, look up your partner in your contacts, and send the details via a messaging app. For this to work, entities need a transferable representation. Siri displays results in cards that developers can design with their own views and branding.

Internal testing only: Siri Integration Advisor and Siri Evaluations

Since a language model never responds identically twice, Apple provides the appropriate tools. The Siri Integration Advisor in Xcode checks the app’s intents code and suggests improvements. With the new Siri Evaluations framework, conversations can be scripted as scenarios and evaluated: which intents Siri invokes, which entities it finds, which skills it loads, and what the response contains. Apple itself recommends running scenarios multiple times and refining the descriptions until the tests pass reliably.

Distribution is severely restricted during the preview:

  • Testing is only possible with developer-signed builds on a supported device or with internal TestFlight testers.
  • External TestFlight testers and App Store users do not receive any of the Siri AI features, even if the code is already included in the app.
  • Skills cannot be tested in the simulator; they can only be tested on actual hardware.
  • The App Intents API and Siri’s behavior are subject to change during the preview.

Apple demonstrates this in 15 of its own apps

Apple is already using these new capabilities in its own apps, including Calendar, Files, Finder, Journal, Mail, Maps, Messages, Music, Notes, Photos, Preview, Reminders, Safari, Shortcuts, and Weather. For example, you can ask Siri about the café you visited last week or have it archive files that no one has opened in over a year. For developers, these integrations also serve as a reference. To find out what other new features iOS 27.3 has to offer, check out our overview of iOS 27.3, featuring the new Health app and release date.

Siri AI Developer Preview: Key Details

Point Details
Preview Launch Date November 3, 2026
Platforms iOS 27.3, iPadOS 27.3, macOS 27.3, visionOS 27.3 (all beta)
Not included watchOS and tvOS
Interface App Intents with @AppIntent(.tool), Skills as SKILL.md, Spotlight index
Requirements Apple Intelligence-enabled device with a supported language; no new entitlement required
Languages English, French, Japanese, Korean, Portuguese, Spanish
Announced Languages Traditional Chinese, Norwegian, Turkish, Vietnamese
Testing Developer builds and internal TestFlight testers
App Store submission Starting in spring 2027
Availability for users “in a few months,” no specific date
EU Not available on iOS, iPadOS, and watchOS; available on macOS and visionOS
Age restriction Ages 13 and up

Siri AI in the EU: iPhone and iPad Left Out

For users in Germany, the situation remains unsatisfactory. Siri AI is still not available on iOS, iPadOS, and watchOS in the EU. Apple merely states that it is working constructively with the EU to find a way forward. The backdrop is the Digital Markets Act: Apple argues that the law would require granting other assistants direct access to private data and the ability to control third-party apps. According to Apple, the European Commission has rejected a proposal for a so-called “Trusted System Agent” with an 18-month transition period. There is no timeline.

On the Mac and Apple Vision Pro, however, Siri AI has been running in the EU since September 14, provided the device is set to a supported language. And that’s exactly where the second catch lies: German is neither on the current list nor among the announced languages. Developers from the DACH region are therefore most likely to be able to test their integrations on a Mac with English as the system language. This is our assessment based on availability; Apple itself has not commented on this specifically.

Incidentally, tvOS is also missing from the preview, even though Siri AI has now arrived on the new Apple TV 4K (2026) with the A19 chip.

Daily Limits and Paid Expansion

One point that almost gets lost in the announcement: Apple Intelligence features that run on server models are subject to daily usage limits. According to Apple, this applies to Siri AI, the intelligent image editing tools, Image Playground, and the AFM-3 cloud models in Shortcuts, among others. The limits are expected to vary depending on the feature, the complexity of the request, and system load. Apple plans to offer expanded quotas for a fee in the future. Apple hasn’t provided specific numbers or prices, and it also remains unclear whether actions in third-party apps will count toward this quota.

Analysis: The Right Move, but with Many Unanswered Questions

Technically, this is the opening that developers have been waiting for since WWDC. Only when Siri AI supports more than just Apple’s own apps and a handful of schema categories will the assistant become truly useful in everyday life. The approach using Tool Intents and Skills in the Agent Skills format is pleasantly close to what has already become standard in the AI industry. Anyone with experience using agents will quickly find their way around. Incidentally, Apple is also exploring other avenues in parallel: With the Creator Studio update, Keynote, Pages, and Numbers are opening up to external AI agents via MCP.

The timeline, however, significantly dampens the excitement. App Store submissions won’t be accepted until spring, Apple hasn’t even published the terms of service yet, and the daily limits remain a mystery. So developers are building on a platform whose rules they don’t yet know. For us in Germany, there’s a double hurdle: no Siri AI on the iPhone and no German language support. Anyone developing apps here in Germany should still get started now. Good intent descriptions and properly indexed content take time, and at least on the Mac and in international markets, this will pay off as soon as Apple flips the switch.

Sources

Categories News Tags Apple, Artificial intelligence, Siri

« Previous articleHyperOS 4 Worldwide: These 82 Xiaomi, Redmi, and Poco Devices Will Receive the Update

Next article »Digital Driver’s License to Launch in Late November on the i-Kfz App

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *