Flat vs Silo

A handy scanner for all sort of 2D and 1D code. Does not require any authentication or internet connection.

Silo runs AI models like Gemma 4, Phi-4, and Mistral directly on your iPhone. No servers. No subscriptions. No internet connection. Your conversations never exist anywhere but your device. Actually Private, Not "Cloud Private" Most AI apps encrypt your data before sending it to their servers. Silo never sends it at all. Zero network requests. There is no server to breach. No account required. No login. No data collection. Anonymous by design. Works Without Internet Airplane mode, off-grid, no Wi-Fi. Silo works everywhere because everything runs on your hardware. No connection means no data leaks. Ever. Chat on a plane, in the subway, or anywhere you want absolute privacy. No Subscription. No Limits. No $20/month plans. No lifetime unlock fees. No usage limits. No accounts. No paywalls. Download and use every feature immediately. Open-Source Silo's complete source code is public on GitHub. Don't take our word for privacy. Read the code and verify it yourself. No trust required. Built by an independent developer, not a corporation. Run Popular AI Models On-Device Run the latest open-source LLMs locally with full Metal GPU acceleration. Gemma 4, Phi-4, Mistral, and more running on your iPhone with no cloud required: - Google Gemma 4 E2B. The newest architecture from Google, available in Q4 and Q8 quantizations. Run Gemma 4 entirely on-device. - Microsoft Phi-4 Mini. Top-performing small model, beats Llama 3.2 on every benchmark. Runs on 4 GB RAM. Phi-4 is the best model for its size. - Mistral / Ministral 3B. Instruct and Reasoning variants from Mistral AI. - Liquid AI LFM 2.5. Ultra-efficient 1.2B models with thinking mode. - SmolLM3 3B. Compact, multilingual, 6-language support from HuggingFace. - Meta Llama 3.2. Fast 1B and 3B models from Meta. - Plus Dolphin uncensored, Phi-4, and more community models. Download the models you want and switch between them. Q4 quantizations for every device, Q8 for devices with more memory. Bring Your Own Models from Hugging Face Import any GGUF model from Hugging Face directly in the app. Paste a Hugging Face URL and download. Run Phi, Dolphin, StarCoder, TinyLlama, or any GGUF-compatible model from the Hugging Face model hub. Your models, your choice. Uncensored Models Silo does not filter, censor, or restrict model output. Use uncensored or custom fine-tuned models with zero guardrails imposed by the app. The model answers. We don't interfere. Run Dolphin uncensored, unfiltered Llama, or any model you choose. Chat History & Conversations Save, browse, and manage multiple conversations. Pick up where you left off. Delete conversations you no longer need. All conversation history stored locally on your device. Nothing is ever uploaded. Streaming Markdown Rendering Responses render with proper formatting as they stream. Bold, italic, headers, code blocks, numbered lists, and nested bullets. No raw markdown syntax cluttering your chat. Fast on Apple Silicon Built with llama.cpp and optimized with Metal GPU acceleration and BF16 compute on supported hardware. Responses stream in real time with zero server latency. Tuned for iPhone performance. Gemma 4, Phi-4, Mistral, and Llama all run smoothly on recent iPhones. Model Management Browse recommended models including Gemma 4, Phi-4, Mistral, Llama, SmolLM3, and LFM. Download with progress tracking and manage your model library. See model sizes before downloading. Cancel downloads anytime. Corrupt file detection keeps your library clean. Your AI assistant should answer to you, not to a cloud provider.