Apple opens Siri AI to all developers for custom app actions

Expanding Siri's capabilities

Apple has released a Developer Preview that allows any app creator to plug their software into Siri AI. Previously, Siri could only perform actions within third-party apps if those tasks fell into specific categories that Apple had already defined. This new update removes those boundaries, allowing apps to teach the assistant their own unique functions.

Under the new system, a developer can enable actions such as checking a delivery status or booking a restaurant table. This means users can ask for these specific tasks in their own words rather than following a rigid set of commands. Apple describes this change as a way to help businesses and developers reach more customers by making apps discoverable through voice requests.

The update is currently available in the betas for iOS 27.3, iPadOS 27.3, macOS 27.3, and visionOS 27.3. While developers can begin testing these integrations immediately, the functionality will not be available to the general public for several months. Apple plans to open App Store submissions for these features in the spring of next year.

How the new framework works

The technical foundation for these changes relies on a framework called App Intents. This allows developers to describe their app's actions, known as intents, and its content objects, known as entities, to system features like Siri and Spotlight. Developers can use a specific Swift macro to mark an intent as a tool, which the AI can then decide to invoke based on the user's request.

To help the AI understand these tools, developers must provide clear titles and descriptions for every action and parameter. Because the model chooses between different tools, the specific vocabulary used by the developer carries significant weight in how the assistant interprets a command. For example, a shopping app might describe an 'add-to-cart' intent so the AI knows exactly when to trigger it during a conversation.

Privacy is baked into this new architecture through various permissions. Developers can require the device to authenticate before Siri reveals certain information, such as a bank account balance. They can also mark specific properties as internal, which completely hides that data from Siri and other system features. This ensures that while the AI is more capable, it only accesses what the developer explicitly allows.

Introducing AI skills

For more complex requests that involve multiple steps, Apple is introducing a new concept called skills. While a tool intent performs a single action, a skill provides natural-language guidance to help Siri combine several actions into one experience. These skills are written in Markdown files and provide instructions on how to handle specific domains, such as completing a full purchase.

A skill file includes a name and a description that Siri uses to decide if the skill applies to a user's request. Once matched, the assistant reads the body of the file, which can contain up to 10,000 characters of instructions. These instructions tell the AI which tools to prefer, what to avoid, and what specific knowledge the app requires to complete the task successfully.

The format for these skills is based on an open standard originally developed by Anthropic for its Claude models. Apple's version is a simplified subset of that standard, focusing on the name, description, and Markdown body. Because skills are essentially advice rather than hard-coded programs, the AI still makes the final decision on how to execute the steps during a live conversation.

Content and context awareness

Siri AI can now find and surface any content that an app places into the device's on-device Spotlight index. This is a significant shift from the previous system, which only allowed the assistant to surface content that matched Apple's predefined templates. By using the Spotlight index as a semantic layer, the assistant can find a wider variety of information across different apps.

The system also uses onscreen context to link what a user is looking at to the entities within an app. This allows a person to refer to a photo or a specific item on their screen conversationally. Additionally, a feature called donations allows apps to report user actions to Apple Intelligence. This helps the system spot routines, such as checking the weather every morning, to suggest actions proactively.

When a request involves multiple apps, the system relies on entities that offer a transferable representation. For instance, a user could ask to email a flight from an airline app to a partner in a messaging app. The assistant identifies the necessary data from the first app and carries it into the second. If the app is in the background, the user must mention the app by name to trigger these cross-app actions.

Testing and rollout timeline

Because AI responses can be unpredictable and vary between runs, Apple has provided new tools to help developers test their integrations. A Siri Integration Advisor in Xcode reviews the code and suggests improvements. Additionally, a new Siri Evaluations framework allows developers to script specific conversations and measure how accurately Siri calls the correct intents and finds the right entities.

During this preview period, the functionality is restricted to development-signed builds and internal testers in TestFlight. Users who download apps from the App Store will not see these Siri AI features yet, even if the app already contains the necessary code. Apple has stated that the goal of this preview is to give developers time to explore and shape the experience before the official launch.

The official rollout to the App Store is expected in the spring, which means these features will not be available for the 2026 holiday season. Apple has not yet published the specific terms and usage policies that will govern what developers are allowed to build. For now, the focus remains on providing the infrastructure for apps to become more interactive with the assistant.