Skip to main content

How your assistant recalls shared memories

Search reaches your open shared spaces, attributed to the person who shared each memory.

A
Written by Aaron Wittman

When you ask a connected AI tool something, its search (search_memories) covers your own vault and memories other members shared into your open spaces. Each shared hit carries a shared block naming who shared it and through which space, and whether you can edit it: Owners and Editors of that space can correct it, Viewers can only read it.

  • "What is our home insurance excess, and who saved it?" — answered from your partner's shared memory, attributed to them.

  • "What's in the Rivera Household space?" — your AI tool reads the whole space with list_space_memories.

  • "Pull up the policy PDF" — the attached document is fetched over the authenticated connection.

Limits, stated plainly

  • Coverage per search is capped (up to 8 member vaults, 25 hits each); if your sharing footprint is bigger than one search can cover, the response says so rather than hiding the gap.

  • Topic context (get_context), the ChatGPT search tool and your core profile draw on your own vault only.

  • Closed spaces are not searched.

  • ChatGPT and Codex never show memories filed under credentials, identity, health or finance, memories holding a password, card or ID number, or memories JustOnce hasn't confidently filed yet — your own or shared ones.

The web app shows a space's contents, but not yet a per-memory "added by" label. Today, who shared a memory is visible to AI tools only.

Did this answer your question?