Siri Is Getting a Better Memory
Apple is building a personal memory system that could help Siri retain information about users' lives, based on new code findings shared by code explorer pdfu . In Apple's iOS 27.3 beta release, the system includes a "memory graph" that's designed to organize information about relationships, places, routines, health, workouts, and what motivates users or holds them back, according to the…
Apple is developing a personal memory system that could enable Siri to retain information about users' lives, based on code insights shared by code explorer pdfu. The new system, introduced in iOS 27.3 beta, features a "memory graph" designed to organize data on relationships, places, routines, health, workouts, and user motivations or obstacles. It also identifies "memory atoms" for facts, opinions, intentions, and events, including details on people, places, times, and topics.
The memory graph employs two processes to gather information. One transforms structured data into insights, like reading name fields in a user's contact card. The other utilizes an on-device AI model or Private Cloud Compute to extract insights by querying sources such as Siri conversations, messages, photos, and health data. Currently, only the AI-driven process for identifying motivations, barriers, and resources using built-in questions is present. These questions include inquiries about routines that might interfere with adequate sleep.
However, this system is still incomplete, with no process yet in place to create memory atoms. The memory graph is distinct from a separate store for explicit memories, which currently has no space for records in Apple's search system. Both systems reportedly use Apple's version of the Kuzu graph database, a technology acquired by Apple in 2025.
The purpose and timeline of Apple's plans for this technology remain unclear, and it is unknown when the unfinished memory system will become available and how users will control their data.
Written by urgent.news from MacRumors's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.