Enclavi User Guide & Architecture
Practical workflows, model configuration, and architectural best practices for your iPhone.
Getting Started & Core Concepts
Enclavi is built on a clear premise: Capture once, shape into a structured working document, and continue thinking without cloud dependencies.
Permissions & Sandboxed Storage
On first launch, Enclavi requests microphone access and optional FaceID authentication. All audio recordings, transcripts, and notes reside solely within your device's isolated application sandbox.
Starting Your First Recording
Tap the microphone icon on the note canvas. Enclavi visualizes the live audio waveform and runs real-time speech transcription via CoreML directly on the Apple Neural Engine.
Refining & Structuring the Document
Extract key highlights from the transcript into formatted blocks, add action item checkboxes, and link key points to exact second-level audio timestamps.
WhisperKit Transcription Profiles & Hardware
Enclavi leverages Apple's 16-core Neural Engine and CoreML execution providers to run OpenAI Whisper models with exceptional battery efficiency.
| Profile | Model Size | RAM Overhead | Speed Factor | Recommended Workload |
|---|---|---|---|---|
| Efficient (Tiny) | ~75 MB | < 180 MB | 12x Realtime | Quick voice memos, on-the-go dictation, minimal battery drain |
| Balanced (Base) | ~140 MB | < 290 MB | 8x Realtime | Standard default for interviews, team standups, and daily notes |
| Enhanced (Small) | ~460 MB | < 620 MB | 4x Realtime | Bilingual meetings, technical vocabulary, and accurate punctuation |
| Turbo (Large v3) | ~1.5 GB | < 1.4 GB | 2x Realtime | Noisy conference recordings on iPhone 15 Pro and later devices |
Bring Your Own AI on Local Network (LAN Ollama & LM Studio)
When you need heavy 70B reasoning models, connect Enclavi to your Mac or desktop PC over local Wi-Fi without sending a single byte to the cloud.
Start Local Server on Mac
Open Terminal on your Mac and launch Ollama bound to all local interfaces using the command above.
Select Endpoint in Enclavi
Navigate to Settings > AI Providers in Enclavi. The app discovers your Mac automatically via Bonjour (mDNS), or you can input your LAN IP address manually.
Zero Public Egress Verification
Test with Airplane Mode enabled and Wi-Fi active. All prompts and synthesis stream directly between your iPhone and Mac.
Document-First Canvas & Source Linking
The document is the permanent deliverable. Audio serves as indisputable evidence and context.
Bidirectional Audio Timestamps
Every sentence in the transcript is tied to the waveform. Tap any word to scrub and listen to the exact audio second immediately.
Excerpt Insertion
Highlight transcript sections and bring them into your note canvas as formatted quotes or actionable tasks.
Rich Multi-Asset Attachments
Attach PDF documents, whiteboard photos, and diagrams. Everything is indexed locally for contextual search.
Ask Your Notes (Local RAG)
Enclavi indexes your thoughts using Apple Accelerate (vDSP) vector embeddings directly on your device. When asking questions in the Ask tab, every response provides direct clickable citations showing note titles, creation dates, and audio timestamps.
Security & Apple Secure Enclave
All personal records and vector embeddings are encrypted at rest with Curve25519 and AES-GCM-256. Cryptographic keys remain isolated inside Apple's hardware Secure Enclave. The automated App-Shutter blurs screens instantly upon app switching and requires FaceID or device passcode to unlock.
Export, Backup & Data Ownership
Export individual notes or your entire archive into clean Markdown (.md) with YAML frontmatter, formatted PDFs, or original audio files (.m4a). Fully interoperable with Obsidian, Notion, Logseq, and Apple Notes.
Frequently Asked Questions
Direct answers to common configuration and technical inquiries.
How do I re-enable microphone permissions if denied?
Open iOS Settings > Privacy & Security > Microphone, and toggle the switch for Enclavi to on.
My Mac is not auto-detected on the local Wi-Fi
Ensure Ollama is running with OLLAMA_HOST=0.0.0.0:11434 and that your Mac firewall permits incoming connections on port 11434.
How much storage do downloaded Whisper models occupy?
The default Balanced model uses approximately 140 MB. You can switch or delete downloaded model files anytime in Settings.