Your AI can reason about almost anything. It just can't remember the file you showed it yesterday. So we built the memory. Index your files once, query them forever. No hardware hassles. Your data is always safe and private.
“The deposit of £4,800 covers scaffolding hire and the first materials order, refundable against the final invoice.”
A drive is a filing cabinet, not a brain. It holds the record perfectly and can still answer nothing about it — because nothing ever read it. That's true of one laptop, and true of twenty years of shared drives.
Kept, backed up, filed under a name someone chose once — and never understood by anything since. In organisations, 55–90% of what's stored is dark data. Your own drive works the same way; nobody has ever bothered to count it.
Every new chat starts with no idea what you've already made — so you find the file, paste it in, explain yourself again, and do it all again tomorrow. At company scale it shows up in the results: 95% of company AI pilots show no measurable financial return, and 85% of the failures come down to the state of the data rather than the model.
Dark data — term coined by Gartner; 55–90% range per Splunk/Gartner surveys, 2019–2025. Pilot failure rates: MIT NANDA, 300+ enterprise deployments, 2025; data-readiness attribution from the same study. Both are organisation-scale studies — the same failure shows up on a single drive, just unmeasured.
Everything you need is technically right there — on a drive, in an inbox, in a repo, somewhere in a two-hour recording. You could find it. You just can’t face it, so it stays where it is. Find and Seek already looked, so you don’t have to: you ask, it hands back the one thing. For you first. Then your AI asks the same way.
Fifteen years of quotes, invoices, photos and receipts you can't face going into. Ask by clause, who was involved, amount, date, or what someone said — get the exact page, cell or timecode.
You find the file, paste it in, explain yourself again. Connect it over MCP via secure API link and it asks the memory itself — a short cited passage instead of five whole PDFs.
Ask a question of everything the company has ever made. People and AI get the exact source, not the whole drive. Each department, case or client sits behind its own wall.
Your files are the room full of books. Searching by filename reads the spines. Searching the text opens every book again, every single time you ask. What you actually want is the librarian who read them all years ago — you ask a question, you get the page. That takes three things a search box hasn’t got.
Not later, when you’re mid-question and sending the whole file over again, but when it first arrives. Read it properly, work out what it means, and keep that.
The people, amounts and dates, and how they relate — tied back to the real document, not replaced by a simplified map of it. You keep the flesh, not just the skeleton.
Reach for one thread and the related material comes back with it — short enough to think about, with the full thing one step away when the summary isn’t enough.
It gets read and indexed once — what it is, who and what it mentions, how it connects to everything else, and where each of those sits in the original. That reading is kept. So when you or your AI ask something later, back comes a card: the passage that answers it, the file and page it came from, and how sure the match is. If the card isn't enough, the full file is one step away. Your files are yours and only yours — zero data retention, never trained on your IP.
Everything else rents you understanding one question at a time. We build it once and you keep it. What a card contains, when your AI asks for more, and where it falls short are written up in full as the Local Insight Cards standard, built on the Hippo Campus Protocol.
A short card is the default, not a limit. Sometimes an answer only shows up if you read the whole file, and a card can miss it. Your AI can always ask for the full thing — but short is what it gets first, and that trade-off is real.
Counting is where we lose. Ask how many of something exists across everything you own and an ordinary text search still beats us, unless your AI picks the counting mode we ship — and right now nothing tells it to. That’s a gap at our end, and it’s on the list.
The hard work happens before you ever ask. Find and Seek understands everything you've got — once — and quietly keeps up as it changes. After that, finding is instant: for you, and for your AI.
Docs, email, code, images, video, calls. Each one is indexed once — what it is, what it mentions, how it connects — and where all of that sits in the original. Your files are never stored or used for AI model training.
One secure link to Claude, Copilot, Cursor, or your own agent. We deliberately don't supply the model — you keep whichever one you like. The memory is the part you own. We never train on your data.
In plain English. Back comes the exact line, page, cell, or timecode — with its source — instead of the whole pile.
Index your files once, securely in the cloud. Your data stays safe and private, forever. No hidden fees, no recurring token lookup costs. Prices in Australian dollars.
Wake up your historical archives in the cloud and stop digging.
Your live daily workstream and persistent second brain.
For massive project backlogs and research archives.
Audit-ready private cloud for teams that need full control.
The beta is open, and you're welcome in it. Your email is all we need to add you to the list — tell us your use case and we'll fit the invite to your setup.
What’s left is scale: making it run instantly on any machine without needing local hardware or model downloads. That’s what the beta is for.