Offline Private AI Library

Offline Private AI Library lets you run modern large language models completely offline on your device. Browse curated MLX-compatible Hugging Face repos, download models with pause/resume, and chat in a clean SwiftUI workspace—no accounts, no telemetry, and no internet required after install.
Why offline?
• Your prompts, responses, and custom prompts never leave your device.
• Works in airplane mode once a model is downloaded to disk.
• Perfect for sensitive research, travel, or teams with strict data policies.
Download and manage MLX models
• Curated Qwen, Llama, Phi, and coder variants grouped by use case, with size and language indicators.
• Save your own Hugging Face repo IDs or open Hugging Face in-app to grab new ones.
• Resume interrupted downloads and get reminder alerts to keep the app open during large installs.
• Delete models with a single swipe to reclaim storage.
Chat smarter
• Markdown conversation view with headings, lists, and code blocks.
• Quick suggestion starters to kick off new chats.
• Share or export conversations to Markdown with timestamps.
• Stop and resume streaming while a response is generating.
• Copy individual replies and delete specific rows from history.
• Inline status banners show when a model is loading or busy.
• Optional safety disclaimer to keep expectations grounded.
Tune it your way
• Theme and accent color customization.
• Prompt profiles for reusable system and starter prompts.
• Forced language: Auto, English, or Russian.
• Controls for temperature, top-p, max tokens, context window, and repetition guard.
• Reset parameters back to defaults at any time.
On-device processing
• Powered by MLX and LocalLLMClient for local inference.
• After download, all model runs stay offline—no cloud calls.
• System info view shows RAM usage and entitlement status so you know which models fit your device.
Offline Private AI Library is a tool for experimentation and learning. It is not a replacement for professional advice and may produce incorrect or outdated information.