Omnesis Vision

Everything you do leaves a trail — emails, messages, files, calendar events, notes, photos, bank transactions, even the little things your phone quietly records about your health. All of it is scattered across a dozen apps and platforms that don’t talk to each other, and nothing you own can see across them.

What exists today

The first thing Omnesis does is fix that. It plugs into the sources that actually matter in your life, keeps them in sync, and indexes everything quietly in the background — on your own machine. One search engine for your entire digital life. One SQL query engine for all your structured data — health, finance, workouts, and the rest. The bar I am setting for myself is that your digital life’s index should fit on a 16GB Mac mini.

Today’s AI agents can reach some of this data through MCP connectors, but there is no connector for your iMessage and WhatsApp history, your Apple Health data, or the web pages you have read. And an agent that doesn’t know where an answer lives has to ask every connector it has, burning latency and quota along the way. An agent pointed at Omnesis has one place to look.

But search alone isn’t enough, because your life doesn’t happen inside one app. A single story — a trip, a project, a friendship — usually plays out across a WhatsApp thread, a few documents in Drive, a calendar invite, a handful of photos. The same goes for the people in it. So Omnesis works out the connections between each piece of data and builds a graph:

• Omnesis knows when a person in a WhatsApp conversation is the same person as the recipient of an email you sent.
• Omnesis knows that a contract you uploaded to Google Drive was shared in that same email thread.
• Omnesis knows that you have an almost identical copy of that contract somewhere else in your Drive.
• Omnesis knows that a landscaping company’s web page in its index was opened because your husband sent you the link on iMessage.
• Omnesis knows that you decided to call that company after visiting their site, because a Granola transcript records a call to a phone number that appeared on that page.

With search alone, a request about “gardening” might turn up the visited web page and the contract, but miss all the adjacent context in WhatsApp and iMessage — and never piece together the timeline of what actually happened. The graph is what makes that adjacent context cheap to retrieve.

That’s the foundation. On top of it sits the Omnesis agent, which you can ask about the most private corners of your digital life. It runs against your indexed corpus on your own Mac mini, and the local databases can be encrypted at rest. If you enable cloud inference, I recommend a zero-data-retention provider. The built-in agent has no web navigation or search: it can answer from your corpus, but it has no way to push anything out of it. If you’d rather, you can plug in your own agent — Hermes, OpenClaw, Claude — and get the benefit of Omnesis with something that can actually act on the world. How much that agent gets to see is up to you. At one end, Direct mode hands it raw retrieval over the sources you tick and nothing else. At the other, you write a privacy policy in plain language and a privacy reviewer applies it to every answer before it leaves: release it, hold it back, redact the sensitive parts, or escalate to your phone so you decide.

Where it goes

Everything above is already implemented. What interests me now is the leap from a system that stores your context to one that understands it. What I’ve built so far is reactive: it answers when you ask. I want something that tells you what you need to know, exactly when you need to know it. So much of life is made of open loops — decisions you haven’t made, tasks you haven’t handled, commitments you need to meet — and each one holds a little of your attention until it’s closed. Imagine a product that carried that load for you:

• A brief before a concert: when it starts, that you and a friend agreed to meet an hour earlier, and the key details from the event page you’d already read.
• A reminder about the email from your lawyer you still need to answer — the one asking for three documents — along with the news that two of them are already in your Google Drive and the third you’ll have to request from your accountant.
• A heads-up to renew your passport before it expires.
• A gentle nudge to call your mum, because you usually call once a week and it’s been ten days.
• A prompt to follow up with your previous landlord, because a week ago he promised your deposit back within two business days and it still doesn’t show in your bank records.

This is what the Omnesis Brain is for: a layer that works quietly in the background to make sense of what the index holds, and that only ever reasons from what your data actually recorded. It is early. The data foundation underneath it — sync, index, graph — is the part that is solid, and it is open source. I would like a community to form around it, building their own algorithms and experiments on top of a substrate they don’t have to rebuild first. The contributing guide is the place to start.