> ## Documentation Index
> Fetch the complete documentation index at: https://docs.praxis-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# How Memory Handles Your Data
> What Memory stores, where it lives, who can see it, how it's used, and how to delete every bit of it.
Memory makes Pria more helpful by carrying your context across conversations. Because that context is *about you*, this page answers the data questions directly: what is stored, where, who can see it, what it is used for — and what it is never used for.
This page covers **memory parameters** ([User Memory](/mdx/user-guide/memory/user-memory) and [Shared Memory](/mdx/user-guide/memory/shared-memory)). Your conversation transcripts, files, and other personal data categories are covered in [Your Data & Privacy](/mdx/user-guide/profile-settings/personal-data).
## What does Memory store about me?
Memory stores **only what emerges from your own conversations** — either because you asked Pria to remember something, or because Pria noticed a durable fact worth keeping (your role, preferences, ongoing projects). Each fact is stored as a small, named parameter with a value and a description — not as raw conversation transcripts.
Memory is not designed to infer hidden traits about you: it records what you said, not conclusions about you.
**Think before you store sensitive details.** Memory will faithfully keep whatever you ask it to — including things like health conditions or dietary restrictions if you choose to share them (they can genuinely help, e.g. *"remember I'm allergic to peanuts"*). Never store credentials, banking details, government ID numbers, or anyone else's personal information. If Pria auto-saved something you'd rather it hadn't, delete that parameter — see below.
## Where is my memory stored, and who can see it?
Memory parameters are stored in Pria's own database, inside your organization's environment — they are not held by any third-party memory service. Who can see a parameter depends on its scope:
| Scope | Created when | Visible to |
| ---------------- | ------------------------------------------------------------------------------------ | ------------------------------------------------------- |
| **Personal** | Chatting outside any Digital Twin | Only you |
| **Twin-private** | Chatting inside a Digital Twin | Only you, within that twin |
| **Shared** | Explicitly saved to the twin's [Shared Memory](/mdx/user-guide/memory/shared-memory) | Every member of that twin (your authorship is recorded) |
Other users can never see your personal or twin-private parameters.
Two administration facts, in the spirit of full transparency:
* **Your organization's Admins** can view and manage members' memory parameters as part of user administration (the same way they can administer accounts and files). See [Memory Parameters & Act As](/mdx/admin-guide/memory-and-impersonation) in the Admin Guide.
* If an Admin uses **Act As** to reproduce a support issue in your account, that session is time-limited and fully audited — the admin's identity, the reason, and the timestamps are recorded in an append-only log.
## How is my memory used?
At answer time, your memory parameters are added to the context Pria's AI model receives — alongside your message and conversation history — so responses arrive already knowing your role, preferences, and projects. That is the whole job.
Your memory is **not** used for advertising, **not** used to build marketing or behavioral profiles, and **never sold or shared** with third parties.
## Is my memory used to train AI models?
**Praxis AI does not train AI models on your memory.** Memory exists to personalize your answers, not to build training datasets.
Like your messages, memory context is processed at answer time by the AI provider your organization has configured (Pria supports multiple providers, including organization-supplied keys — see [Bring Your Own Token](/mdx/introduction/byot)). It is sent only as conversation context for generating your answer, exactly as your typed message is.
## How long is my memory kept?
Memory parameters are kept **until you (or an Admin) delete them** — there is no automatic expiry, which is what makes memory useful weeks or months later.
Deletion is immediate and permanent:
* Deleting a parameter **hard-deletes** it — it is removed from the database, not flagged or archived. There is no undo.
* **Delete everything at once** from **Your Profile → Your Data → Delete my data → Personal memory**. Pria starts learning about you fresh from your next conversation.
* Your memory entries are also included when you [export all your data](/mdx/user-guide/profile-settings/personal-data#exporting-all-your-data-as-a-zip) as a ZIP.
## How do I view, edit, or delete individual memories?
You have full read/write control, from two places:
Ask *"What do you remember about me?"*, correct a fact (*"Actually, my role changed"*), or say *"Forget what I told you about X"* — Pria updates or hard-deletes the parameter.
**Your Profile → Memory** lists every parameter with its value and description. Edit or delete any row directly. What you see here is exactly what Pria knows.
See [User Memory](/mdx/user-guide/memory/user-memory) for the full management walkthrough.
## Can I turn Memory off?
Yes. The **Enable User Memory** switch (on by default) lives in two places: **Your Profile → Knowledge → Memory**, and as a quick toggle in the conversation **action menu → Knowledge → Memory** (where *Manage Memory…* appears while memory is on). Turning it off means Pria stops saving new personal memories and stops using the ones it already has — in every conversation, across text and voice.
Three things stay true with the switch off:
* **Nothing is deleted.** Your existing parameters are kept (and shown in the Memory panel) — they're just not used. Flip the switch back on and personalization resumes where it left off. To erase entries, use the Memory panel or **Your Data → Delete my data**.
* **Shared instance memory keeps working.** Your team's shared knowledge is organizational content, not personal data — it is unaffected by your personal switch.
* **You can still be selective instead.** Tell Pria not to remember something in the moment (*"don't remember this"*), or delete individual parameters — the switch is for turning the whole capability off.
**Twin builders:** you can also disable User Memory per assistant (the *Disable User Memory* option in the assistant's settings), so conversations with that assistant neither read nor write memory parameters regardless of each user's switch.
## Related
What Pria remembers about you and how to manage it
Team-wide knowledge inside a Digital Twin
Inspect, export, and delete everything Pria stores about you
How Admins administer memory and audited impersonation