The trillion-token life — and who should hold it
In May 2025, on stage at Sequoia’s AI Ascent, Sam Altman was asked how ChatGPT becomes truly personal. His answer skipped past features and named the destination: the ideal, he said, is “a trillion tokens of context that you put your whole life into.” Every conversation you’ve ever had. Every book you’ve read, every email, everything you’ve ever looked at — in the context window, with your life appending to it as you live it. And, he added, your company doing the same with its documents.
Read it twice, because it is not a claim about chatbots. It is a claim about your documents. The most powerful person in AI is saying that the model is not the prize — the corpus is. The thing that makes AI useful to you, specifically, is everything you have written, received, signed, and saved. The reasoning engine is becoming interchangeable. Your archive is not.
We think he is exactly right about the destination. We built this company on the same belief.
The question he left open
Notice what the trillion-token picture quietly assumes: all of it lives inside one platform. Your life becomes their context. The design has real consequences, and they don’t depend on anyone behaving badly — they follow from the architecture itself.
A profile that accumulates. A context that keeps appending is, by definition, a permanent and growing record of you, held by someone else. The more useful it gets, the more complete it becomes.
Leverage that points the wrong way. If the accumulated context is what makes the AI good, then whoever holds the context holds the switching cost. Your own life becomes the reason you can’t leave.
A single blast radius. One account — compromised, subpoenaed, or simply repurposed by a future terms-of-service — is no longer an inbox. It is the searchable index of everything you are.
None of this is an argument against the destination. It is an argument about custody.
The same destination, held by you
There is another way to build the trillion-token life: the corpus stays yours. Your documents sit in your library — statements, contracts, notes, research, the decades of paper a business or a life produces. When you ask a question, the AI reads them at question time, never to train on — nothing becomes training data, and inputs and outputs are automatically deleted from the AI provider’s systems within 30 days. We’ve stated plainly what happens to your documents. Every answer carries citations back to your own pages, so you can check it in seconds.
You get the part of Altman’s vision that matters — an AI that reasons across everything you know — without the part that should give you pause. The archive compounds. The profile doesn’t.
You don’t have to wait for it
Altman was careful to call his trillion-token model an ideal that can’t be reached today. True — for a model that holds your whole life in working memory. But for the corpus you already own, the useful version exists now: upload it, ask it anything, verify every answer against your own pages. The reading you were never going to do, done — and the record of your life still filed where it belongs. With you.
Read nothing. Know everything. — docustrata.com
DocuStrata is not affiliated with or endorsed by Sam Altman, OpenAI, or Sequoia Capital; the quoted remark is from his public appearance at Sequoia’s AI Ascent in May 2025, and the application of it here is ours.