Spotlight: minimi — a memory layer that watches everything, keeps it local
Most AI memory tools ask you to feed them things: paste a doc, connect an integration, tag a conversation as important. minimi flips that. It sits in the background on your Mac, watches what you read, say, and hear across apps like WhatsApp, Gmail, and Calendar, and tries to surface the things you actually meant to follow up on but didn't. The pitch is right there in the tagline: an ambient memory layer for Claude.
The problem it's aimed at is a real one. If you run your work life through Claude and a scatter of apps, the actual list of commitments you've made lives nowhere. It's in a WhatsApp reply you sent at 11pm, a calendar invite you half-read, an email thread you meant to close out. minimi's job is to reconstruct that list, which it calls "open loops," and then, more ambitiously, to close some of them without you doing anything.
The smart decision here isn't the memory idea itself, plenty of tools chase that. It's the packaging: minimi is built specifically as an MCP for Claude, not a standalone AI assistant competing for your attention, and it stores everything on-device with no cloud database. That second part matters more than it might first appear. A tool that promises to capture "everything you have ever read, said or heard on your Mac" is a hard sell if you don't trust where that data goes. By keeping it local and folding into Claude rather than replacing it, minimi is betting that the wedge is trust plus convenience, not another chat window.
Who should try this: Claude users on a Mac who already lean on it as a daily driver and who feel the tax of context-switching between email, chat, and calendar. If you're the kind of person who ends the day having said yes to five things across three apps and remembers none of them by morning, this is squarely built for you. The "works right out of the box, no integrations needed" claim is also worth noting if true for your stack, since setup friction kills most memory and productivity tools before anyone gets to see the value.
Who should not bother: anyone not on a Mac, obviously, since there's no other platform mentioned. Anyone not already using Claude, since this isn't a general-purpose memory tool, it's an accessory to one specific assistant. Teams looking for shared or collaborative memory should also look elsewhere. This is explicitly personal and single-device: "your memory that you own," stored on one machine, for one person. And if the idea of an app passively reading your WhatsApp and Gmail activity makes you uneasy even when the data never leaves your laptop, the privacy architecture won't fully solve that discomfort, it just changes where the trust question lives from "do I trust their servers" to "do I trust this app running locally."
The part I'd want to see more of is what "closes them automatically" actually means in practice. Tracing a decision and surfacing it to you is one thing. Autonomously resolving a loop, presumably by drafting or sending something on your behalf, is a different level of trust to ask for, and the website doesn't spell out where that line sits. That's the detail worth digging into before you hand it real commitments.
Where this could go next, and this is a read rather than anything they've stated: the obvious expansion is beyond Mac and beyond Claude, since the open-loops idea isn't inherently tied to either. The harder and more interesting path is doubling down on the local-only architecture as the actual differentiator, since "AI that remembers everything but never leaves your device" is a positioning that gets more valuable as more competitors default to the cloud.
Try minimi: projectminimi.com
See the launch: minimi on welaunch.sh