Skip to main content
Your digital twin uses four complementary memory systems. Understanding the difference helps you get consistent, personalized answers without paying for context you don’t need. Two of them — Long-term Memory and Shared Memory — are together called Twin Memory: what your twin carries between conversations.

Conversation History

Short-term context for the current conversation

User Memory

Facts and preferences you save by hand

Long-term Memory

What your twin picks up on its own, as you talk

Shared Memory

Knowledge shared with everyone using the same Digital Twin

Conversation History — short-term context

Your digital twin’s “working memory” for the current conversation: recent exchanges are included with each new prompt so it can follow what you’re talking about.
  • Temporary — only the current conversation; resets when you start a new one.
  • Adjustable — the Remember History slider (sidebar → Knowledge) runs from 0 to 15 dialogues; default is 3.
  • Cost-aware — more history = more tokens per request. Once a conversation runs past five exchanges, smart compaction summarizes the older ones automatically (unless your twin’s administrator has switched compaction off).

User Memory — long-term personalization

Persistent memory parameters (key/value facts with descriptions) Pria keeps about you: your role, preferences, projects, and style. Stored automatically when something important comes up, or explicitly when you say “remember that…”. You can view, edit, and delete every parameter from the Memory panel (avatar menu → Memory).
Full guide: User Memory

Long-term Memory — what your twin keeps on its own

Where User Memory holds the facts you save, Long-term Memory holds what your twin notices for itself. When something in a conversation genuinely changes how it should help you later, it keeps that — no “save this” button, no prompt from you. If you do ask (“remember that I prefer metric units”) it saves that too, and confirms it. Most exchanges leave nothing behind: it keeps what matters, not a transcript.
  • You always see it happen. Whenever an exchange touches long-term memory, a Memory row appears under the answer, in the same expandable style as the Search Results row. Open it to see each memory involved, tagged Recalled, Saved, Updated or Forgotten, and click one to read exactly what was kept. Anything confidential is listed without its content until you click Reveal.
  • Yours alone. A long-term memory belongs to you on that Digital Twin. Nobody else sees it, and administrators see counts only — never the content.
  • You stay in charge. Browse, search, correct, export or permanently delete everything it has kept in Settings → Memory (new interface) or Manage Memory… → Long-term Memory (classic interface, from the [+] action menu’s Knowledge submenu — see Classic interface / New interface if you’re not sure which you have). The same place lets you choose whether your twin builds memories on its own or sticks to the facts you save by hand.
Long-term Memory is additional — User Memory and Shared Memory keep working exactly as before. It is switched on for a Digital Twin by the Praxis AI team (its own administrators see the switch greyed out); if you don’t see a Long-term Memory section under Settings → Memory, it isn’t enabled for you yet.
Full guide: Twin Memory — receipts, confidential memories, managing and exporting, and how it is enabled.

Shared Memory — team knowledge

A Digital Twin’s collective bulletin board: parameters shared with every member of the twin. Use it for team standards, policies, and reference information that should guide everyone’s conversations. Twin admins add and maintain shared entries; every member’s conversations benefit from them automatically.
Full guide: Shared Memory

Comparison


Best practices

  1. Adjust history length to the conversation — high for deep work, low for quick questions.
  2. Move durable facts into User Memory — preferences stated once in memory beat re-explaining them in every chat.
  3. Let Long-term Memory do the noticing — where it is enabled, just talk; open the Memory row when it appears to check what was kept, and correct it in Settings → Memory if it got something wrong. Say “confidential” when something must stay private.
  4. Use Shared Memory for team standards — set them once, everyone benefits.
  5. Review periodically — clear stale memory parameters, outdated shared entries and long-term memories you no longer want so personalization stays accurate.