Equinox AI is an on-device, privacy-first multimodal assistant built
using Apple Foundation Models, Vision, Speech, and
SwiftUI.
It seamlessly combines text, voice, and camera/image inputs
to create a fast, intelligent, and fully local AI experience --- no
cloud, no servers, no data leakage.
⚡️ All inference runs entirely on-device using Apple Intelligence--powered Foundation Models.
- Powered by
FoundationModels - Low-latency responses
- Works offline (once models are ready)
- No network calls, no cloud inference
- Live speech recognition
- Tap-to-record interaction
- Auto-transfers spoken queries into the chat prompt
- Natural conversational flow
- Capture images directly from the camera
- OCR using Vision framework
- Converts visual content into text for LLM reasoning
- Fully local Vision + ML pipeline
- No external APIs
- No image uploads
- No voice data stored remotely
- Built entirely on Apple's on-device ML stack
User Input
├── Text (SwiftUI)
├── Voice (Speech Framework)
└── Image (Camera + Vision)
↓
Multimodal Prompt Builder
↓
Apple Foundation Language Model
↓
Response
- SwiftUI
- FoundationModels
- Vision
- Speech
- UIKit
- Swift Concurrency (async/await)
- iOS device that supports Apple Intelligence
- iOS 18+
- Xcode 16+
- Physical device required
- Clone the repo
- Open in Xcode
- Select a real device
- Build & run
- Multimodal memory
- Visual question answering
- Scene understanding
- Persistent on-device knowledge
Shreyas Battula
MS Computer Science @ UCR
AI / ML / On-Device Intelligence