While You Sleep, Local AI Read All Your Messages

Rohan Paul's commentary: Local models have reached the point where it's worth having a model read through your entire digital life every night.

Sentient OS (YC F26) officially launched today. It runs on the Mac and uses a local LLM to read your messages, files, and notes at night. Every night, it writes what it learns into a folder of plain Markdown files on your computer. You can open, edit, or delete them.

Launch video.

During the day, Double Tap and Sidekick read these files to help you draft replies and complete tasks in apps. Double Tap: double-click the right ⌘ key, and the AI pastes a reply draft directly into the input box based on your context and writing style. Founder Aditya Vellanki says this solves the pain of waking up every morning to 15 emails.

Double Tap demo: double-click ⌘ to generate a reply.

Another experience is computer use. Click the notch on the screen and say “handle this for me,” and the AI can take over operations on the screen. Vellanki believes this can extend computer use to the remaining 80% of your daily life where you haven't used it yet. Vellanki says the unlimited inference enabled by local LLMs is the foundation for both features.

Computer use demo: click the notch, and the AI takes over the screen.

Vellanki emphasizes ownership: “You own all of Sentient's knowledge. It's a Markdown folder on your Mac, written by an AI you control.” It currently has more than 10,000 users and over 130 enterprise customers, including Adobe, Nebius, and Liquid AI. Open source. Website: sentient-os.ai

Knowledge base demo: personal Markdown memory files generated by AI. In the demo, the AI generated a structured knowledge base for a person: categories such as career and recognition, company, and educational background, all organized in Markdown.

The product splits “AI reads all your data” into two steps: nightly batch processing uses a local model; during the day, lightweight calls read local memory; only heavy work requests a cloud model. Data stays on the device, and control stays with the user.

But the discussion quickly turned to concerns. Some users online said local operation is the entire selling point. No one would let a cloud model read their iMessage every night, but running on-device is appealing. Some pointed out that the Markdown folder will become a new core asset: once an email with a prompt injection is summarized into memory, every subsequent AI answer could be contaminated. Each note should have a source and a simple way to remove it.

Others asked about legal risk in work settings: what happens if AI notes on a company computer are subpoenaed? Some also cared about the actual experience: how much battery does a local model use if it runs all night? One user joked: Claude could do this too, but if you make it read all your emails, it might not like you.

Sentient's trade-off is to keep data on the device and hand management responsibility to the user as well. The Markdown in the folder can be deleted, but only if you remember to look at it.