Last week we opened up Memory Vaults to everyone. It was a big week of fire drills. Upgrading servers, storage, and constant performance fixes. The response was overwhelming.

We saw over 40,000 accounts now have a personal data vault they control, linked to at least one social account like Twitter or Farcaster.

This being crypto we know that “unique addresses” is not always the most reliable metric, but every one accounts being linked to a unique Twitter or Farcaster link makes us confident that a large number are real human accounts with multiple verified social links.

It begins to strengthen our hypothesis: The demand for personal sovereign data stores is being pulled forward due to the need for LLM personalization.

Memory Vaults are the main consumer product for Memory.

It’s a digital data vault for everyone to control and manage their data, earn from it, and take it with them between apps and LLMs. We’re rolling it out in chapters.

• Curate their digital identity in one place (wallets, Twitter, Github etc.)

• Earn $MEM balance for linking their accounts.

• Receive real revenue share when their identity data is queried via our API.

• Upload your LLM history, take control of your “memory,” and port it between LLMs.

• Connect external apps (like Spotify and [what else??]) and upload personal data into your vault for richer personal context.

We originally aimed to have 100k Memory Vaults live by the end of the year, but we are already on track to surpass that.

The demand for a personal, sovereign data store has been accelerated dramatically by the conversation around LLM memory. In the last two weeks alone:

The likely answer is an open data layer where the user controls their memory, and can take it with them between LLMs. The conversation is already heading to this conclusion:

Ashley Mayer (GP Coalition Ventures)  - “Imagine an external layer where apps could write to and read from your memory with permission. If there’s anything I want to own/control (for professional + personal use), it’s this.”

Aaron Levie (CEO Box) - “‘Plaid for memory’ is a very good concept. Hard to pull off, but memory portability and ways of providing new AI systems instant context will be extremely useful in the future.”

There is only a small window of opportunity to build this user-owned memory layer (about 18 months) before it closes. And that means the winner of this category likely already exists.