Sentient-OS-Labs/sentient-os

Sentient OS – an on‑device personal AI for macOS

What it is – Sentient OS is an open‑source macOS application that acts as a personal AI “butler”. It watches the files, emails, calendars and other sources you let it read, builds a knowledge base stored as ordinary Markdown files, and then lets you hand off small tasks to the AI. The AI can type, click, fill forms, draft replies, and even prepare proactive suggestions while you sleep.

Key capabilities

• Sidekick – invoke the assistant with a shortcut (right‑Command). It gets its own cursor and can operate your apps (email, browser, forms, etc.) while you keep using the computer.
• Double‑Tap – double‑tap the shortcut while the cursor is in a reply field; the assistant drafts a message in your voice using the on‑screen conversation, your knowledge base and a one‑time set of writing examples.
• Proactive intelligence – overnight the app reads enabled sources, summarizes them with a local model, and surfaces “cards” in the morning (e.g., pending replies, subscription renewals, travel plans).
• On‑device inference – a built‑in lightweight model runs on Apple‑silicon GPUs (Metal) for the heavy‑volume reading. You can swap in a larger local model (e.g., Qwen 3.8‑27B) or use any compatible hosted model (ChatGPT, Claude, OpenRouter, etc.) for the reasoning steps.
• Privacy‑first design – raw files never leave the Mac unless you explicitly enable cloud sharing. Knowledge is stored in ~/Sentient OS - Knowledge Base/ as plain Markdown. Optional encrypted sync uses AES‑256‑GCM with a one‑time secret link.

Installation

brew install --cask sentient-os-labs/tap/sentient-os

• Requires Apple silicon (M1+), macOS 15.4+, ~8 GB RAM for the built‑in model and ~10 GB free disk for the initial setup.
• A pre‑built binary is also available on the GitHub releases page.

How it works under the hood

• Engine: a customized LiteRT‑LM integration runs Gemma 4 E4B on the GPU with speculative decoding, visual‑token budgeting, and per‑item context isolation.
• Pipeline: selected local sources → on‑device summarization → optional cloud or local frontier model → knowledge folder → UI actions (Sidekick/Double‑Tap).
• Scheduling: a macOS‑level overnight scheduler wakes a plugged‑in Mac (lid closed), processes sources, then returns the machine to sleep.

User‑controlled data

• Enable/disable any source (Mail, Calendar, Apple Notes, iMessage, WhatsApp, files, screenshots) in Settings.
• Edit the generated writingstyle.md to shape the tone of drafted replies.
• View, edit, or delete the knowledge base directly in Finder or via the Knowledge window.
• Turn on/off optional encrypted cloud sharing; when enabled, only encrypted markdown files are uploaded, and the decryption key never persists on the server.

Business model

• The app itself is free for the first 1,000 users (AGPL‑3.0 license). Paid AI providers (OpenAI, Anthropic, etc.) are billed separately if you choose a hosted model. The company plans to monetize through enterprise integrations and support.

Where to learn more

• Demo video: https://sentient‑os.ai/
• Privacy policy & technical data‑flow: https://sentient‑os.ai/privacy and SECURITY.md
• Engine source code: Sentient OS macOS/Engine/Engine.swift
• Scheduling details: Sentient OS macOS/Scheduling/Documentation - Overnight Scheduler & Wake Helper.md

Bottom line – Sentient OS is a genuine, open‑source AI‑powered personal assistant that runs primarily on your Mac, giving you hands‑free task automation while keeping your data under your control.