iOS 27.3 reveals Siri's memory graph: memory atoms fed by Fastpass and agent recipes
alexisgallagher · x · 2026-11-07
A developer has surfaced details of Siri's AI personal-assistant memory system in iOS 27.3: a memory graph centered on the user, covering relationships, places, routines, health, workouts, and what motivates or holds them back.
Its basic units are "memory atoms" — facts, opinions, intentions, and events — each tagged with time, place, people, and topic. The graph is filled by two kinds of recipes:
- Fastpass: converts already-structured data into insights, e.g. Person-v1 reads name fields from the Me card;
- Agent: an on-device or Private Cloud Compute model extracts insights by querying Siri chats, messages, photos, and health data. Today there is a single agent recipe, MotivatorBarrierResource, whose prompt includes 60 built-in questions.
The poster's takeaway: we'll see innumerable AI harnesses for the same reason we have innumerable apps — the harness is how you customize AI capability for a specific product.
More from coding & agent
- Google DeepMind's Ivan Leo to reveal how agents cheat on benchmarks at scale at AIE Code — jocarrasqueira · 2026-11-07
- Sudo mode verification revamp: passkey-first UX, inbox deep links, paste capture — JohnPhamous · 2026-11-07
- Amp teases a new way to interact with agents — vinvan · 2026-11-07
- Cryptographers flag that AI agents fail Cloudflare challenges; blind signatures may fix it — jedisct1 · 2026-11-07
- Vercel Tracing goes GA on all plans, enabled by default in Next.js 16.5 — cramforce · 2026-11-07
- Decompiling Battlefield 1942 with Claude Code costs ~$100 in API tokens — alvelda · 2026-11-07