- Siri Pro is an on-device large language model and multimodal AI engine for iOS, iPadOS, macOS, watchOS, and VisionOS.
- New A19 Bionic and M4 chips power Siri Pro’s local processing for enhanced privacy and speed.
- AgentKit enables third-party apps to integrate Siri Pro’s capabilities for complex, cross-app tasks.
Apple used WWDC 2026 to do something it has been promising for years: put a genuinely capable AI engine on the device itself. Siri Pro is the result, a large language model built to run on the A19 Bionic and M4 chips, handling complex queries and multi-step tasks without routing your personal data to the cloud.
Siri’s next-generation AI capabilities are the result — a large language model built to run on Apple’s latest chips, handling complex queries and multi-step tasks without routing your personal data to the cloud.
Siri Pro Powers On-Device Intelligence Across Apple Ecosystem
The headline claim from the keynote is privacy through architecture. Because Siri Pro’s core intelligence runs on-device, Apple says most interactions never leave your phone. That matters in practice: responses are faster, and you are not handing over your Calendar entries or Messages to a remote server every time you ask a question.
The demos made the ambition clear. Siri Pro drafted emails from a few spoken prompts, summarised long web articles and planned travel itineraries by pulling from Calendar, Maps and Messages simultaneously, all without the user manually opening each app. Apple says the same experience extends to watchOS and VisionOS, not just iPhone and Mac.
Apple says the same experience extends to iPhone and Mac.
What has changed most visibly is conversational memory. Earlier versions of Siri required you to restate context with every new question. Siri Pro is designed to hold the thread across an extended back-and-forth, understanding follow-up questions as part of the same task rather than isolated commands.
AgentKit Opens Siri Pro to Developers
The more consequential announcement for how your apps actually behave may be AgentKit, a new framework that lets third-party developers plug into Siri Pro’s capabilities. Apple’s Vice President of Software Engineering described how it works: developers define specific “skills” for their apps, and Siri Pro can invoke those skills on a user’s behalf, according to Apple. That shifts the relationship between Siri and apps from “voice shortcut” to something closer to a background assistant that acts across them.
Two examples Apple gave show the range. An airline app could automatically rebook a flight when a calendar event shows a conflict. A fitness app could adjust your training plan based on health data and a casual spoken goal. Neither requires the user to navigate to the app manually.
Apple built a sandboxing layer into AgentKit so third-party agents can only access user data with explicit permission. Each app’s skills are isolated from others, meaning a travel app cannot read your fitness data unless you say it can. That privacy architecture appears to be a direct response to the concerns that have followed AI assistant launches elsewhere, as seen with incidents where AI agents acting with excessive access caused unintended consequences.
New Hardware Accelerates On-Device AI Performance
None of this works without the underlying chips. The A19 Bionic, destined for upcoming iPhones, and the M4, already in current Macs and iPads, both feature a rebuilt Neural Engine Apple says delivers roughly twice the computational throughput of the previous generation, according to the company. That headroom is what makes running a large language model locally viable rather than theoretical.
The redesign is not just about raw speed. Apple engineered specialised memory architecture into the new Neural Engine specifically to handle the large parameter counts that trip up on-device AI, reducing the bottleneck that has made local AI sluggish on older hardware. The payoff shows up in features like real-time voice generation, image analysis inside the Photos app and predictive text that learns your patterns without syncing that data off the device.
The broader strategy here is Apple’s familiar playbook: control the chip and the software together, and you get performance no third-party combination can easily match. Whether Siri Pro delivers on that promise in daily use remains to be seen, but the architectural pieces, dedicated silicon, on-device processing and a developer framework to build on top, are now in place. For more on how AI is running closer to the edge, see our coverage of multimodal AI running locally on consumer hardware. Explore more AI tools and tips in our Consumer AI section.
