Your context isn't supposed to sit somewhere.
It's supposed to show up.
Today, every conversation starts with a tax. You pay it in tokens, while the model rediscovers what could already be known. You pay it in quality, because nothing was prepared for the question: your commitments, decisions, and preferences sit scattered across sources instead of one source of truth. And you pay it in time, re-explaining instead of working, watching an indicator think, think some more, and almost finish thinking.
Fabric exists to end that tax: the right context, already chosen, already there.
Not an archive you search. A stream that arrives already chosen.
Streamed, not stored.
Fabric pushes the exact slice you need into the tool you're already working in, the moment you need it. No archive to query, no prompt-wrestling, no starting from zero.
Chosen, not dumped.
The hard part was never moving context. It's knowing which slice fits which model for this task. Fabric keeps your commitments, preferences, and risks distinct, never mashed into a paragraph, and continuously chooses what to stream. Only the slice the task needs ever leaves.
Yours, into any model.
One context you own, that follows you into Claude, ChatGPT, Cursor, whatever you reason in. Your store stays yours; what leaves is the slice, to the model you chose, and nothing else.
Quality, then cost, then time. In that order.One person already runs his whole working life on it.
The founder has run every meeting, commitment, and decision through Fabric, daily, since March 2026. He won't work without it. Read that for exactly what it is: proof the mechanism works, and an honest admission that you'd be early. He runs a heavier setup than you will. What Fabric does for someone who simply works across meetings, docs, and more than one AI tool is what the next handful of customers are for.
Works with what you already use.
Every source is handed over on your own screen: your mail arrives as an export you download yourself, and nothing is uploaded anywhere. Something else in your stack? Say so in the get-started form and you're first in line when it lands.
One thing to install. Free while it's early.
Your stream on your own machine, inside Claude Desktop. It puts the Claude subscription you already pay for to work: Fabric adds the context, not a second AI bill. The file is a free download, and it builds your store with you, right in the chat.
Self-serve, on purpose. Install it in a two-minute drag into Claude Desktop, then say set up my fabric stream: it builds your store right there in the chat, from an export of your own mail, on your own machine. Company-name invoicing is ready for when pricing turns on.
No login, no blank screen. It works from the first minute and says so.
The honest answers.
What is Fabric, exactly?
A context store of your own, in your own database, that streams the right slice of your work into the model you're working in. Today that's Claude Desktop, where Fabric lives as a free extension; the store itself is model-neutral, built to brief whatever you reason in next. Everything on offer today is free and you install it yourself, including your full store: say 'set up my fabric stream' inside Claude and the extension builds it right there in the chat, from an export of your own mail, on your own machine. It tells you honestly where it stands at every step.
Is my client-confidential context safe?
Your context lives in a store of your own: in the self-serve setup, a database on your own machine, never pooled with anyone else's. The extension reads nothing until you hand it an export of your own mail, and nothing is uploaded anywhere. When you work, only the slice a task needs leaves the store, to the model you chose. Minimization isn't a policy bolted on; it's how it's built.
Do you support the tools I use?
Today Fabric streams into Claude Desktop, as a free extension. Your store is model-neutral by design; ChatGPT, Cursor, and anything that speaks MCP are the doors planned next. On the source side it reads your mail today (Gmail via a Takeout export; Outlook and Thunderbird via the collector kit), with Google Calendar, meeting transcripts, working documents, Drive sync, and Slack next. Something else in your stack? Say so in the get-started form and you're first in line when it lands.
Which email does it read?
Gmail today, through a Google Takeout export you download yourself: the extension builds your store from it, on your own machine, and nothing is uploaded anywhere. Outlook, Microsoft 365, and Thunderbird mailboxes go through the free collector kit instead. Something else in your world? Tell us in the get-started form and you're first in line when it lands.
Do I have to change the tools I already use?
No. Fabric briefs the tools you already open. It doesn't ask you to move into a new app. Keep your stack; it just stops starting from zero.
How fresh is it?
Your store covers up to the date of your last mail export. Drop a fresh export any time to move it forward; repeats are deduplicated, so it's always safe. Live sync is next on the roadmap, and this page will say so plainly when it lands.
My ChatGPT already has memory.
For ChatGPT. Switch to Claude for reasoning or Cursor for code and you're back to zero. Fabric isn't another assistant with memory. It's the layer that briefs the assistant you're in: Claude today, with a store built to carry the same context anywhere next.
Is it really free?
Yes. The Desktop Extension is a free download that reports its own state honestly before any store exists, and building your store with it is free while it's early. There's nothing to pay yet, and nothing here asks for a card.
Stop starting from zero.
One free extension. Your stream on your own machine, on the Claude you already pay for.
No login, no blank screen. Every piece works from the first minute and says so.