Sotto – On Device AI

Sotto is a chat app for language models that run on your own device. There is no account, no server and no analytics. Your conversations stay in the app’s own container on your Mac, iPhone or iPad. TWO WAYS TO RUN A MODEL • Apple’s on-device foundation model. Nothing to download; start chatting straight away on a device that supports it. • Open-source models. Download a GGUF model from a curated catalog, or import a .gguf file you already have. Sotto runs it with llama.cpp, using Metal on device. Once a model is on the device, Sotto keeps working with the network off. TOOLS THE MODEL CAN CALL Twenty-five built-in tools run entirely on your device: dates and time zones, a calculator and unit converter, percentages and number bases, text rewriting, find-and-replace, extraction, sorting, word counts and diffs, JSON and CSV, descriptive statistics, encoding, hashing, random numbers, and a search of your own past chats. Four are on out of the box; the rest are one switch away. You can add two of your own: a web search that uses your own API key, or an HTTPS request you define. Both stay off until you set them up, and each asks for approval before it runs. MAC, IPHONE AND IPAD • A menu bar item on the Mac: ask a question from the status bar, reach your last five chats, and open any window in one click. • Compare two models side by side on the same prompt. • Personas: saved system prompts and sampling settings. A local-only persona never sees a tool that uses the network. • Export any conversation to Markdown or JSON. • Optional Face ID, Touch ID or passcode lock. PRIVACY Sotto’s developer collects nothing. No sign-in, no identifier, no usage statistics, no third-party SDK. The only outbound requests are a model download you start, an optional weekly check of the model catalog (off by default), and any tool you set up yourself. Settings › Privacy shows a live count of the bytes the app has actually sent. FREE Sotto is free. No in-app purchases, no subscription, no ads and no paid tier. A NOTE ON GENERATED TEXT Sotto runs models locally and does not filter, review or fact-check what they produce. Models invent things, get facts wrong, and can produce mature language or themes. Treat an answer as a draft and check anything that matters.