Skip to content

Ask Your Vault (optional, bring-your-own-key)

Ask Your Vault is an optional, opt-in feature we are preparing in Labs. It is not yet enabled. This page explains, in plain language, how it works — and what it means for your privacy — so you can understand it before it goes live.

What it does

You ask a question; an AI model answers using memories from your own vault. It works only with an AI API key you supply (your own OpenRouter key). It is read-only — it can never add to, change, or delete anything in your vault.

Where your data goes (and doesn't)

When you ask:

  • The memories you select as scope, plus your question, are sent from our server to OpenRouter, which routes the request onward to the AI provider you chose. Depending on that provider, this may involve the United States or other countries.
  • Zero-data-retention (ZDR) routing is requested by default — the request is directed only to providers that agreed not to retain it. You can turn ZDR off only by knowingly choosing a non-ZDR model, and we warn you first.
  • A secret safety-net scans what you are about to send (and the reply) for anything that looks like an API key, token, or credential. If it finds one, the request is blocked and you are asked to remove it. It is a safety net, not a guarantee.
  • Never sent: your whole vault, another user's data, or memories that are deleted, superseded, archived, stale, or in trash. The model receives temporary labels for your memories, not their permanent identifiers.
  • Your key is held only in memory for the session and is never stored; it is discarded when you log out. No transcripts are kept — our logs record only counts (scope, model, tokens, cost), never the content and never your key.

The honest limit

Once your content is sent on your key to OpenRouter and the provider you picked, we cannot control what they do with it — their terms govern that, not ours. Ask Your Vault is a transfer you initiate, like exporting to a third party of your choosing. We do not claim your data "never leaves your control."

Privacy summary (our data-protection assessment, in plain words)

We wrote a short internal data-protection assessment for this feature. In plain language:

  • The main residual risk, stated first: once you send content on your key, how the destination provider handles it is outside our control. ZDR-by-default reduces this but does not remove it, and turning ZDR off is your explicit choice.
  • Purpose: let you query your own memories (or our public product docs) with an AI model of your choice, on your own key.
  • What is involved: the memories you scope plus your question — which can include personal information you chose to store.
  • Lawful basis and special-category data: these are legal questions we are putting to our solicitor before the feature goes live. We are not self-deciding them here.
  • What we already do to reduce the risk: ZDR by default; the secret safety-net (which blocks, not "cleans up after"); your key session-only and never stored; no transcripts; token limits and no automatic retries; and the answer shows which memories it drew from (citations) so you can judge whether it is grounded in your vault.

See the Privacy Policy for the formal statement, and the Terms of Service for your responsibilities when using it.