vMiraAll posts
product · April 28, 2026 · 2 min read

Cross-chat shared context — Mira remembers what you talked about

A per-profile memory layer with cosine-similarity retrieval and time decay lets Mira recall prior conversations when they are relevant — opt-out in profile settings.

Quick answer

vMira chats now share context across conversations. A per-profile memory layer uses cosine similarity with time decay to surface relevant prior context. Opt-out in profile settings.

Mira now remembers what you talked about across chats.

The mechanism is a per-profile memory layer. When you start a new conversation, the system looks at what you have discussed before and surfaces what is relevant — past projects, ongoing work, recurring preferences. The retrieval is cosine-similarity over an embedding index with a time-decay weight, so recent and salient context wins over stale, off-topic context.

Why this matters

If you have ever explained your tech stack twice, your team structure twice, or your house style twice — to the same assistant — you already know why. Memory is what makes a co-worker different from a temp.

What is and isn't shared

  • Shared: topics, preferences, working context — only within your account.
  • Not shared: other users, public web data, anything outside your account.
  • Opt-out: the profile settings page has a master toggle. Toggle off and the retrieval is skipped.

We will continue to err on the side of less, not more, until we have lived with this for a while. If you find Mira recalling something you would rather it forgot, open the profile memory page and clear it.

memorycontextrelease
By vMira Team
Cross-chat shared context — Mira remembers what you talked about — vMira Blog — vMira