[OS] 3 months ago, 2,500 of you joined the waitlist from my post here. Sentient OS is out! An on-device LLM understands your entire life overnight, and you wake up to Proactive Intelligence: your work already done, one click from firing through Computer Use. Free & open source :D

3 months ago, I posted Sentient OS here, and 2,500 of you joined the waitlist. So many of you wrote to me over these months; I read every single one. It's out right now at https://sentient-os.ai, and it grew into something much bigger than what I'd pitched you. TLDR: An on-device LLM understands your entire life, then proactively offers to get your work done through computer use. Here's how! - Every night at 3 AM, it quietly wakes your Mac (lid closed is fine) and a custom on-device LLM reads what's new in your life: files, screenshots, WhatsApp, iMessage, Apple Notes, email. All of it is read locally on your own chip (raw data never leaves your Mac) and distilled into a clean markdown knowledge base (an Obsidian-style folder that's yours to read and edit). - Proactive Intelligence: you wake up to work Sentient noticed, researched, and prepared on its own, one click from firing through computer use. The reply you forgot, drafted from your rich personal context. The subscription renewing tomorrow, caught tonight. You read it, edit if you like... and nothing ever fires until you click. - Sidekick: computer use that knows your whole life, anywhere on your Mac. Click your notch and just say "finish this for me"; the notch glows with status as it does the task in your own apps and your browser, even in the background while you keep working. And because it's grounded in your knowledge base, it knows what's happening in your life. Bonus: your knowledge base can connect to your ChatGPT & Claude over MCP. Connect it, then ask your ChatGPT "what do you know about me?". Be amazed :] Comparison Siri AI: Cannot click around your Mac and actually DO stuff for you across your apps (only has a few limited things/tool calls it can do). Only has personal context from inside Apple's walled garden, and uses RAG for this instead of an LLM-curated markdown knowledge base (the best personal context for agents). No proactive intelligence! You gotta ask it to (hopefully) get it to help. Sentient understands your entire life and can use any app just like you do. And it looks out for you proactively. And the whole thing is now open source (AGPL): the app and the infrastructure. Every privacy claim above is verifiable in the code: https://github.com/Sentient-OS-Labs/sentient-os Under the hood To build these near magic features, we push the bounds of what's possible with on-device inference. It's the star of the whole show: truly proactive AI has to re-read your entire life, every single day, and that much inference is only free, private, and unlimited on your own chip. So the local LLM (Gemma 4 on a custom inference framework) does ~90% of the compute: it reads everything locally, summarizes, junks the noise, flags what's worth acting on proactively, and strips PII (with deterministic scrubbing behind the model as a backstop). Only those clean summaries ever leave. GPT-5.6 Sol then organizes them into your knowledge base, running through your own ChatGPT subscription via your own Codex CLI on your Mac (Sentient sets this up for you). Pricing Sentient is free forever for non-enterprise users. Your Mac does the compute, so it costs us nothing to run. So how do we make money? Once we refine and iterate on this new frontier with all your feedback, we'll swap out the connectors for Slack, Notion, Granola, etc. and bring Sentient to enterprise. Sentient's consumer version will always be free. Enterprise is where we make money. The full engine (the proactive mornings + the notch agent) currently needs you to have a ChatGPT Plus subscription. A ChatGPT free plan has enough Codex usage to get the knowledge base. And that's just for now: the engine isn't tied to any one provider (Codex CLI is OSS), and we'll be opening up to more AI providers of your choosing in the near future. And if you have a crazy amount of ram (>= 48 gigs since we need a lot of context kv cache), you'll soon be able to run your fav "frontier" local LLM for 100% local instead of 90% local! :) Download: https://sentient-os.ai Source (star the Repo! :D): https://github.com/Sentient-OS-Labs/sentient-os Apple Silicon (M1 or newer), macOS 15+, and 8 GB of RAM is enough (we tuned inference to make it happen!) Trust / transparency I'm Jesai! You might know my open-source Writing Tools (2.3K+ GitHub stars, ~30 press features). Or from my previous post here on Sentient! Since that last post, I'm now in SF with my best friend & co-founder Aditya to build Sentient. https://www.linkedin.com/in/jesai-tarun-37004725a/ Would love to answer everything; and very happy to geek out about the on-device stack (we run a 4B multimodal model with vision on 8 GB Macs via a custom inference fork with MTP, flash attention, & smart KV cache reuse), how we got computer use running on the plain Codex CLI (involves reverse-engineering a DMG), or reverse-engineering macOS to be able to wake your Mac up for inference at 3 AM! To the 2,500 who waited; thank you. This one's for you :D