You say it once.
Over lunch you tell ChatGPT the things that decide every meal in your house: vegetarian, a tree-nut allergy, batch-cooking on Sundays. It listens beautifully. It remembers perfectly. And it tells no one.
So on Wednesday you say it again to Claude. And on Saturday to Gemini. You are the integration layer between your own tools, and it's unpaid work.
It becomes a file you own.
A second after you said it, Suvadu remembered it — as a trace — as markdown, in a bucket in your own Google Cloud project. Not a row in our database. A file, with your name on the bill.
You can open it. Change “vegetarian” to “pescatarian” and every tool changes its mind at once. Delete the line and it's gone everywhere, immediately, with nothing to request and no one to email.
Say something true. Watch it travel.
Type a sentence — or take one of ours — and watch it become a file you'd own, then watch three other tools answer differently because of it. This is an interactive walkthrough: nothing you type is saved, sent to an AI tool, or leaves this page. And one caveat that applies to every example on this site: AI tools can make mistakes. Memory makes them better informed, not infallible — check what matters.
Every tool reads it.
The next morning you ask Claude whether pesto is safe for your daughter. You don't mention the allergy. It already knows, because it read the trace over MCP, the open standard your tools already speak. Joining one is a click, not a project. Five are live today:
And you never open it again.
Sunday, eight people for dinner, and nothing to explain to anyone. Suvadu has no app you open and no notifications to dismiss. It shows up as a sentence you didn't have to type, inside a tool we don't own.
A trace is a trace, dinner or database.
Nothing about the format is domestic. The decision your team argued over on Tuesday is the same kind of file as the tree nut allergy. Cursor, VS Code and Claude Code read it the same way, because they all speak MCP.
Which means the reason you chose the ugly migration survives the sprint, the tab, and the model that suggested it.
We never read it.
Most hosted memory products hold your life on their servers and ask you to trust the policy. Suvadu runs in a Google Cloud project that belongs to you, so there is no copy on our side to be tempted by, subpoenaed for, or lost in a breach. One honest limit: once you let Claude or ChatGPT read a trace, that provider's own data policy applies to what they receive.
We're a family-run studio with no investor waiting for this to become an advertising business. Which raises a fair question, so we'll answer it next.
And if we disappear, your memory keeps working.
You should ask a two-person studio this before trusting it with the shape of your family. So: if we stopped tomorrow, your traces would still be markdown files in a bucket you own, and the server reading them would still be running in your own Google Cloud project.
Nothing of ours sits in the path at read time. There's no key to expire, no API to shut off, and nothing to export, because MCP is public and the files were always yours. What you would lose is us: no more updates, no new tool integrations, and no one to email when a provider changes their config format. The memory survives; the maintenance doesn't.
Getting one takes less time than this page.
A guided wizard provisions into your Google Cloud project while you watch it happen. No terminal, no config file, and nothing left behind if it fails.
One thing to know before you start: Google needs billing enabled on that project, even though the free tier usually costs nothing. That is Google's requirement, not ours — we never see the card and never bill it.
And you pay for it exactly once.
Every AI tool in your life bills you monthly. This one asks once. $49 is about two months of a single AI subscription, and then never again, for every version we ever ship. Start free with 50 traces — no account with us, just your own Google sign-in — and your cloud bill stays between you and Google.