Local capabilities.
Global context.

How it works

384-DIM

Vector Space

Index your Apple data

Gryz reads from your iCloud Drive (Gryz folder), Calendar, Reminders, and Contacts. All indexing happens on-device using Apple's native frameworks.

4096

Context Window

Ask naturally

Chat with Gryz like you would any assistant. Ask about your calendar, notes, contacts, or reminders. "What's coming up this week?"

<2.0s

First Token

Get cited answers

Every response includes source citations — the calendar event, note, or contact that informed the answer. All inference runs locally on your iPhone.

evt_calendarOK

obj_contactsOK

doc_notesOK

Device Boundary

Gryz Engine

A17 Pro / A18 Neural Engine

BLOCKED

Remote Server

NO CONNECTION

Built for extreme privacy.

Everything you need, nothing you don't.

Two-Stage Models

Starter model (~1GB) downloads automatically so the app works instantly. Full model (~2-3GB) available as a free upgrade. No paywall on intelligence.

STARTER_MODELQwen 2.5 1.5B

FULL_MODELQwen 2.5 3B

Source Citations

Every AI response shows exactly which note, calendar event, or contact informed the answer. Full transparency on how the AI reasons.

RETRIEVED CONTEXT

meeting_notes_april.md

evt_dentist_appt

Streaming Chat

Watch the AI think in real-time with token-by-token streaming. Optimized for Metal GPU acceleration.

<2.0s

FIRST TOKEN

On-Device Voice

Push-to-talk input via SFSpeech. Audio never leaves the device.

User Rules

Set persistent instructions. "Always use bullet points." Gryz remembers.

Live Indexing

Watch indexing progress source-by-source. See exactly what Gryz has learned.

The difference is structural.

Not just another chatbot. The only AI with local Apple ecosystem context.

Specification

Gryz

Cloud AI

Inference Environment

On-device (iPhone)

Cloud (Remote servers)

Apple Data Access

Full (Notes, Cal, Contacts)

None (Requires upload)

Account Required

Yes (Email/Phone)

Offline Capability

Full RAG & Chat

None

Network Exfiltration

Zero (Tier 1 Data)

100% (All queries sent)

Authentication

Face ID / Touch ID

Session / Password

Hardware Limits

On-device AI needs modern hardware.
We don't compromise.

A model that runs smoothly on an A17 chip is frustratingly slow on an A16. We set a hard minimum to guarantee a real-time conversational experience.

Minimum

iPhone 15

A17

Processor Architecture

6 GB

Minimum RAM Required

17.0+

iOS Version

Note: iPhone 14 and older lack the Neural Engine throughput required for acceptable inference speed. The app will not install on unsupported devices.

Recommended

iPhone 16

A18 Pro