Privacy
How LucidPal keeps your data entirely on your device.
Fully On-Device
Every word you type, every calendar event you manage, and every AI response is processed locally on your iPhone. Nothing leaves your device.
- No account required
- No API keys for AI inference
- No subscription
- No internet connection needed to use the AI (after initial model download)
What LucidPal Accesses
| Data | Used for | Leaves device? |
|---|---|---|
| Your messages | On-device AI inference | Never |
| Calendar events | Reading and writing via EventKit | Never |
| Photos / images (optional) | On-device vision analysis | Never |
| Microphone (optional) | Speech-to-text recognition | Never |
| Contacts (optional) | Looking up contacts via Siri | Never |
| Notifications (optional) | Reminders and alerts | Never |
| Device RAM | Selecting the right AI model | Never |
Calendar Access
LucidPal requests access to your iOS calendar via EventKit — the same system framework used by the built-in Calendar app. You control which calendars are accessible through iOS Settings.
To grant or revoke access: iOS Settings → Privacy & Security → Calendars → LucidPal
Speech Recognition & Voice Recording
When you speak to the Agent, LucidPal records audio locally using the iOS AVAudioRecorder API and transcribes it on-device via WhisperKit (a local implementation of OpenAI Whisper). This audio:
- Stays on-device — it is never uploaded to any server.
- Is discarded after transcription — the raw audio file is deleted once your speech has been converted to text.
- Can be disabled by revoking microphone permission in your device settings. Note that doing so will prevent voice input in the Agent.
To grant or revoke microphone access: iOS Settings → Privacy & Security → Microphone → LucidPal
Microphone access is optional. The app works fully without it.
Conversation History
Your chat sessions are saved locally in the app's private storage on your device. They are:
- Not backed up to iCloud by default
- Not synced across devices
- Deleted permanently when you delete a session or the app
The AI Model
The Qwen3.5 model files are downloaded once from HuggingFace (huggingface.co) and stored locally on your device. Once downloaded, LucidPal works completely offline — no internet connection is needed for AI inference. The model weights are never sent anywhere — all inference runs via llama.cpp with Metal GPU acceleration directly on your iPhone's Neural Engine.
Vision and Photo Analysis
If you use the optional vision model, any photos or images you share are processed entirely on-device by the vision model. Images are resized and passed to the local model for analysis — they are never uploaded to any server.
To enable vision: select a vision model during onboarding or in Settings → Vision Model. See the Models guide for how to download and configure vision models.
Web Search (Optional)
Web search is off by default. When enabled, LucidPal supports three providers:
| Provider | What is sent externally | To whom |
|---|---|---|
| DuckDuckGo (default) | Search query only | DuckDuckGo (html.duckduckgo.com) |
| Brave Search | Search query + your API key | Brave (api.search.brave.com) |
| SearXNG | Search query only | Your own self-hosted instance |
Your messages, calendar data, and conversation history are never included in search requests — only the query text extracted by the AI is sent.
The Brave API key is stored only on your device (iOS Keychain via UserDefaults). It is never transmitted to Anthropic or to LucidPal servers.
lucidpal.app Web Services (Optional)
A separate cloud backend (lucidpal-api) powers optional cloud AI features. It handles account authentication and billing for subscribers.
All core iOS features run fully offline. Calendar management, on-device AI inference, and voice input never connect to any server. Cloud AI requests (when enabled) are routed through the LucidPal backend — your conversation content is never stored server-side.
No Analytics
LucidPal collects no analytics, no crash reports, and no usage data. There is no telemetry of any kind.