LucidPal
On-device AI calendar assistant for iOS — powered by Qwen3, with optional cloud AI via Gemini.
What is LucidPal?
LucidPal is an on-device AI assistant that understands and manages your iOS calendar through natural language. Every word is processed locally on your iPhone — no data ever leaves the device.
| Feature | Description |
|---|---|
| 100% On-Device | All inference runs via llama.cpp on the Neural Engine. No API keys required for local AI. Optional cloud AI available with subscription. |
| Cloud AI | Optional cloud AI powered by Gemini 2.5 Flash. Available for Starter and higher subscribers. Automatically switches to local when offline. |
| Calendar Integration | Create, update, delete, reschedule, and query events using plain English. Conflict detection included. |
| Siri Shortcuts | Ten built-in Siri intents let you manage your calendar, notes, contacts, and reminders without ever opening the app. |
| Multi-Session | Full conversation history with named sessions, persisted locally across launches. |
| Notes | Capture quick thoughts in chat; notes are enriched and stored locally. See the Notes guide. |
| Vision / Photo Analysis | Describe images and photos using an optional on-device vision model. See the Models guide. |
| Proactive AI | Daily morning briefing notification via background refresh. Tap to launch Agent Mode. |
| Premium Tiers | Starter, Pro, and Ultimate plans with cloud AI credits, live activities, smart widgets, and more. |
| Live Activities | Dynamic Island and Lock Screen Live Activity during AI generation. |
| Widgets & Notifications | Home-screen widgets and pre-event push notifications. See the Widgets guide. |
| Contacts | Look up phone numbers and email addresses from your contacts in chat. |
| Reminders | Set one-off reminders through natural language — synced to the iOS Reminders app. |
| Web Search | Query DuckDuckGo, Brave, or a self-hosted SearXNG instance — results synthesized locally. |
| Conversation Templates | Built-in AI personas (Writing Coach, Decision Helper, Meeting Prep, Brainstorm) for focused sessions. |
| Productivity Features | AI actions for notes from chat, conversation export, full-text message search, and pinned prompts. |
| Settings | Full reference for every setting in the app. See the Settings section within the app. |
| Privacy | Full explanation of the on-device architecture and what data never leaves your phone. See the Privacy guide. |
| Models | How to download, switch, and manage GGUF models from the app. See the Models guide. |
Quick Start
- Download LucidPal from the App Store and open it.
- Download an AI model — pick the recommended model for your device on the "Choose Your AI" screen and tap Download & Get Started.
- Grant calendar permission — LucidPal will ask during onboarding so it can read and write your events.
- Start chatting — type or tap the mic and ask anything: "What do I have tomorrow?", "Schedule a dentist appointment Friday at 3 pm", or "Clear my Tuesday afternoon."
First-Run Experience
On first launch, LucidPal walks you through five onboarding screens:
| Step | Screen | What happens |
|---|---|---|
| 1 | Your Pocket AI | Overview of on-device, no-cloud design |
| 2 | Knows Your Schedule | Introduction to calendar integration |
| 3 | Type or Speak | Mic and text input introduction |
| 4 | Choose Your AI | Select a text model (required) and an optional vision model, then download |
| 5 | Data Sources | Enable Notes, Contacts, Calendar, Location, and Web Search. iOS permission prompts for Calendar, Contacts, and Location appear inline so you can grant access in context before entering the app. |
After step 5, you are taken directly into the app with all selected permissions granted.
Device Requirements
LucidPal runs the AI model on-device via llama.cpp with Metal GPU acceleration and supports all iPhones with 2 GB+ RAM:
| Device RAM | Supported model | Status |
|---|---|---|
| 2–3 GB | Qwen3.5 0.8B | ✅ Supported |
| 3–5 GB | Qwen3.5 2B | ✅ Supported |
| 5 GB+ | Qwen3.5 4B or Vision 4B | ✅ Recommended |
The minimum iOS version is iOS 16 (for AppIntents/Siri Shortcuts support). iOS 17 or later is recommended for the best experience.
Models
LucidPal ships with four built-in GGUF models, automatically selected based on device RAM:
| Model | Size | Min RAM | Capabilities |
|---|---|---|---|
| Qwen3.5 0.8B | 0.51 GB | 2 GB | Text only |
| Qwen3.5 2B | 1.2 GB | 3 GB | Text only |
| Qwen3.5 4B | 2.5 GB | 5 GB | Text only |
| Qwen3.5 4B Vision | 2.5 GB | 5 GB | Text + Vision (integrated) |
See the Models guide for full details.
Tech Stack
| Layer | Technology |
|---|---|
| Language | Swift 5.10 + SwiftUI |
| LLM Runtime | llama.cpp (C FFI via Swift actor) |
| Calendar | EventKit (wrapped in CalendarService) |
| Speech | AVFoundation + SFSpeechRecognizer |
| Vision | llama.cpp multimodal (optional GGUF) |
| Siri | AppIntents framework |
| Persistence | Custom JSON session store |
| Testing | XCTest with full mock infrastructure |