Infinity Vaults is built almost entirely on Apple’s own frameworks, running advanced processing on the device across artificial intelligence, media, vision, and security, with no dependence on external servers.

On-device intelligence

• AI assistant: local LLM inference through MLX, GGUF models via llama.cpp, and Apple’s Foundation Models (iOS 26+), running on the Apple Neural Engine and GPU (Metal) with streaming, tool calling, persistent memories, and on-device tokenization. On iOS 27 the assistant adopts Apple’s Core AI inference layer, including an instant zero-download tier powered by the built-in system model and FLUX.2 Klein image generation.
• Private AI server: the assistant can connect to the user’s own self-hosted Ollama server over the local network, keeping large-model inference on hardware the user controls. It can also use Apple’s Private Cloud Compute.
• Image generation: Core ML (Apple’s ml-stable-diffusion; SD 1.5/2.x/SDXL) on the Neural Engine and GPU.
• Image Playground: integrates Apple’s Image Playground (ImagePlayground, iOS 18.2+) to generate images from a text prompt, or transform an existing photo with a prompt, entirely on-device.
• Photo intelligence: Vision auto-tagging with 1,300+ on-device categories localized in 16 languages, subject and document detection, one-tap subject lift for cutouts and stickers, and duplicate detection (feature prints).
• Speech: on-device audio transcription via Apple’s SpeechAnalyzer (iOS 26+), turning recordings and podcast episodes into encrypted, editable notes at many times real-time speed.
• System captions (iOS 27): live, on-device generated subtitles during video playback through AVFoundation.
• Text analysis: Natural Language for on-device processing of notes and content.
• On-device audio generation (iOS 27): Stable Audio models run on Apple’s Core AI inference layer to create music and sound effects entirely on the device, from short clips and effects to full-length instrumental tracks of several minutes. Audio agents give the assistant a Generate Audio tool, so a chat can score a scene without any audio leaving the phone.
• Image enhancement: AI Upscale and AI Sharpen run Core ML models through Vision on the Neural Engine and GPU, tiled and blended with Core Image. Users can download additional restoration models, and processing resolution scales with device memory.
• All local; nothing is sent to a server. Optional cloud models (OpenAI, Gemini, and Anthropic chat, plus Gemini image generation) are opt-in and clearly labeled.

iPhone Duo

• Both displays, edge to edge: built with the iOS 27.1 SDK. Layout follows the window scene’s size, so every screen reflows as the device opens, closes, and rotates, and file grids, the browser, and the AI chat use the full inner display.
• Camera regions: controls and toolbars read the system’s reserved regions and move clear of the camera on each display.
• Capture: the in-app camera follows AVKit’s capture direction coordinator, so the correct lens is chosen in each posture.

Capture, media & editing

• Camera & audio: AVFoundation for in-app capture (zoom, exposure) and a rebuilt segment-based audio recorder with live waveforms. iOS 27 deferred capture start roughly halves camera launch time, with smart framing (Auto Zoom / Auto Rotate) being adopted.
• Video pipeline: a full AVFoundation transcoder (HEVC/H.264, resolution, crop, and bitrate control with storage-savings estimates) plus frame-accurate video editing: crop, adjustments, filters, text, and stickers rendered per-frame with Core Image.
• Image editing: Core Image adjustments and filters, PencilKit drawing, and true redaction (blur and pixelation baked into the exported image).
• Audio editing: trim and merge via AVFoundation composition, exporting standard AAC.
• Live Photos: import and play Live Photos (PHLivePhoto).
• Playback: Picture-in-Picture and AirPlay via AVKit, plus background audio with lock-screen and Now Playing controls (AVAudioSession, MediaPlayer) for saved audio and podcasts.
• Import, export & preview: Photos / PhotosUI, PDFKit, QuickLook, and UniformTypeIdentifiers.

Private browsing

• WebKit (WKWebView) full browser with content blocking (40,000+ trackers), in-app downloads, and home-screen web apps. Downloads land directly in the encrypted vault, never the OS’s shared storage.
• Require VPN: an optional setting pauses browsing and downloads whenever no VPN tunnel is present and resumes when it returns. Detection reads the device’s network interfaces, so it works with any provider.
• Network Status: an on-demand check, confirmed by the user each time, shows the address, location, and network the internet sees for the device.

Security & access control

• AES-256 envelope encryption: each vault has its own random master key, with per-purpose keys derived via HMAC-SHA512. Master keys are wrapped by a key derived from the user’s passphrase (PBKDF2, 210,000 iterations) and protected with Security / CryptoKit and the Keychain. No vendor key, no backdoor, no server-side recovery.
• Key lifecycle: in-app encryption upgrade for existing vaults, on-demand key rotation, and an exportable per-vault recovery key (text and QR), backed by redundant keyring copies with self-healing integrity checks at every unlock.
• Streaming encryption: multi-gigabyte files are encrypted and decrypted in constant memory and never staged as plaintext.
• LocalAuthentication (Face ID / Touch ID) and Core Location (geofence) for unlock, plus break-in reporting.
• Self-destruct: optional full erase after a chosen number of failed unlock attempts, layered on escalating lockout backoff. Key material is destroyed first, so all ciphertext is rendered unreadable instantly.
• Key-Bound Encryption: optionally, the keys a user enters (PIN, pattern, or password) become part of the key that unwraps each vault’s master key, so nothing stored on the device can open the vault without them. A vault password becomes a second key for that vault.
• Secure Enclave unlock: for key-bound vaults, Face ID or Touch ID releases a wrapping key held in the Secure Enclave, with no passcode fallback. The key never leaves the device and is never included in a backup.

System & OS integration

Content enters and is protected inside the enclave through deep integration with iOS:

• App Intents: core vault actions (open a vault, search, add a note or file, quick capture) exposed to Spotlight, Shortcuts, and Siri.
• Share Extension: send photos, files, and web content into the encrypted vault directly from any app via the iOS share sheet.
• Home-screen widget (WidgetKit): fast camera capture straight into the vault from the Home Screen.
• Files: import and export through the system document picker (UIDocumentPicker) with full Uniform Type Identifier support.
• Photos & Contacts: import from the system photo library (Photos / PhotosUI) and address book (Contacts) into protected storage.
• Deep linking & continuity: a custom URL scheme and NSUserActivity for navigation and state restoration.
• Reminders: scheduled on-device with UserNotifications, with no calendar or server dependency.
• Subscriptions: StoreKit for IV Professional.
• Desktop access: a built-in WebDAV server over the local network, plus Finder/iTunes File Sharing, for moving files from a computer.
• Cloud vaults: optional vaults stored in Google Drive, or Dropbox, holding ciphertext only, in accounts the user controls.

Privacy architecture

• Content is captured, stored, browsed, and processed in-app. It never passes through the OS’s shared plaintext stores (camera roll, Files) or third-party servers.
• Every intelligence feature (tagging, transcription, generation, chat) runs on-device by default. Anything that leaves the device is explicit and opt-in.
• Cloud vaults write ciphertext only to the user’s own Google Drive, or Dropbox account.

Platforms: iPhone & iPad, iOS 17.6+ (Foundation Models and transcription require iOS 26+; Image Playground requires iOS 18.2+; Core AI, system captions, and smart framing require iOS 27; iPhone Duo support requires iOS 27.1).