Sentient OS, a Y Combinator startup, launched a Mac app on 27 October 2026 that runs a local AI model overnight to read a user’s messages, mail, notes and files. Co-founder Aditya Vellanki announced it on X as “on-device AI that understands your life overnight with a local LLM.”
The company’s site says the app is “Free for life” with a launch code, and its GitHub page says the offer covers the next 1,000 users. The source code is public under the AGPL-3.0 licence.
Short explainer · Sentient OS
A vertical news Short narrated from the published text of this piece and related FourWeekMBA pieces.
Business Pill · On-device AI
An evergreen Business Engineer explainer on on-device AI: running a model on the phone or laptop rather than in the cloud. It explains the general idea and does not discuss Sentient OS.
The key insight: As we read it, Sentient OS is splitting AI work by cost: the GitHub page says the repeated reading of a user’s messages, mail and files runs on the Mac, and that the user chooses the model for the more demanding reasoning.
Vellanki’s launch post says the app’s “unlimited inference enables 2 magical experiences.” The site names them Sidekick and Double Tap.
Sidekick, the site says, “uses your computer like you do with its own cursor” to finish tasks you hand it. Double Tap drafts replies when you double tap your right command key; the site says Sentient “writes what you’d have written, because it knows your life.”
The GitHub page lists the sources the on-device model can read: files, saved screenshots, iMessage, WhatsApp, Apple Notes, Apple Mail and Apple Calendar. It says useful summaries become connected notes about your people, projects, plans and preferences.
Where each step of Sentient OS’s work runs, as the company describes it: a built-in Gemma 4 E4B model reads local sources on the Mac and turns them into summaries; the model the user chooses consolidates them into Markdown knowledge and powers Sidekick; Double Tap’s covered default is GPT 6.1 Sol via the OpenAI API with Zero Data Retention. Sources: Sentient OS GitHub README and privacy page, read 27 Oct 2026.
What Was Posted on X
Today, we’re launching Sentient OS (YC F26), on-device AI that understands your life overnight with a local LLM. This unlimited inference enables 2 magical experiences ๐งต
Aditya Vellanki, a co-founder of Sentient OS, posted on 27 October that the company is launching Sentient OS (YC F26), which the post calls on-device AI that understands your life overnight with a local LLM, and that this unlimited inference enables two experiences. This piece uses the post for the date and wording of the launch.
Local models have reached the point where it makes sense to let one read your entire digital life every night.
Sentient OS just launched an on-device AI that reads your messages, files and notes overnight with a local LLM.
Each night it writes what it learns into a folder of plain markdown files on your computer, which you can open, edit or delete. During the day, Double Tap and Sidekick read from those files to draft your replies and finish tasks in your apps, calling a cloudmodel you choose only for the heavier work.
Rohan Paul posted on 27 October that Sentient OS reads messages, files and notes overnight with a local LLM, writes what it learns into plain Markdown files, and that Double Tap and Sidekick read from those files during the day, calling a cloudmodel the user chooses only for the heavier work. This piece checks that summary against the company’s own site and GitHub page.
What Runs on the Mac and What Does Not
The GitHub page says the local pipeline runs Gemma 4 E4B through a customised LiteRT-LM integration, using the Mac’s GPU through Metal. It lists the requirements as an Apple silicon Mac, M1 or newer, on macOS 15.4 or later, with 8 GB of RAM for the built-in model.
The heavier reasoning goes to a model the user picks. The privacy page lists three options: your own ChatGPT or Claude account, OpenRouter, or a larger local model through LM Studio or Ollama.
Double Tap has its own default. The privacy page says the drafts Sentient covers use GPT 6.1 Sol through the OpenAI API with Zero Data Retention, and that the company covers that cost for its first users. Users can switch Double Tap to another provider or a local model in its settings.
What the app learns is stored as Markdown files on the Mac. Vellanki wrote on X: “You own all of Sentient’s knowledge. It’s a folder of markdown on your Mac, written by AI you control.” Sharing that knowledge with other AIs over MCP is off until the user enables it, the privacy page says.
The Cost Argument
The GitHub page sets out why the reading happens locally: “A tiny handoff needs rich context, and rich context takes a lot of inference.” It says putting that repeated reading on the user’s own chip makes it “practical and free of per-call cloud costs.”
The same page explains how a free app pays its way: “On-device understanding keeps our costs low. Our business is enterprise: work integrations, commercial licensing and support.” It adds that paid AI providers have their own charges.
What’s new, as we read it: a startup is splitting the work by cost, with the high-volume reading done on the user’s own Mac and only the heavier tasks sent to a model the user chooses.
The Structural Read
As we read it, the local part is the high-volume part: the GitHub page says a tiny handoff needs rich context, and that rich context takes a lot of inference.
As we read it, the cloud part has not gone away: the privacy page says Double Tap’s covered default uses GPT 6.1 Sol through the OpenAI API, and Sidekick can use a ChatGPT or Claude account.
As we read it, the free tier rests on that split: the GitHub page says on-device understanding keeps the company’s costs low and that its business is enterprise.
Sentient OS GitHub README, read 27 October 2026
“On-device understanding keeps our costs low. Our business is enterprise: work integrations, commercial licensing and support.”
Three Implications
FOR MAC USERS The GitHub page lists the requirements as an Apple silicon Mac, M1 or newer, on macOS 15.4 or later, with 8 GB of RAM for the built-in model.
FOR MODEL PROVIDERS The privacy page says users can bring their own ChatGPT or Claude account, OpenRouter, or a larger local model through LM Studio or Ollama.
FOR COMPANIES The GitHub page says that for a limited time the first teams get all of Sentient for free, and that the source is licensed under AGPL-3.0.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, Sentient OS moves the highest-volume queries, the overnight reading, onto the user’s own chip, where the company says there is no per-call cloudcost. The framework was written before this launch and does not discuss Sentient OS.
What Is Not Established
The descriptions here are Sentient OS’s own, from its site, GitHub page and posts on X, and we did not test the app. The company has not said how much funding it has raised beyond being backed by Y Combinator, and it does not publish user or revenue figures we could verify. Windows is not available yet.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Sentient OS launched a Mac app that runs Gemma 4 E4B locally to read a user’s messages, mail, notes and files overnight, and keeps what it learns as Markdown files on the Mac. The company offers the app free for life to its next 1,000 users and says its business is enterprise licensing and support.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Descriptions are from Sentient OS’s own site, privacy page and GitHub README, and from posts on X by co-founder Aditya Vellanki and by Rohan Paul on 27 October 2026, all read that day. They are company claims; we did not test the app. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
