When memory is on, PrivateMind remembers durable facts you share in your conversations and uses them in later chats, so you don't have to repeat yourself. Tell it once that you work on credit-risk models or prefer short answers, and it carries that forward. What it remembers is private to you, and you can delete it at any time.
Memory is off by default. Nothing is stored until you turn it on.
Turning memory on
Memory takes two switches: your organisation enables it, then you opt in.
- Your administrator enables Memory for your organisation.
- You turn it on for yourself at Settings → Memory, with the Enable memory toggle.
If the toggle is greyed out with "Memory is turned off for your organization," your org hasn't enabled it yet. Ask your administrator.
Memory is per-user. Turning it on affects only your account, and other people in your organisation never see what you've stored.
What gets remembered
PrivateMind only forms memories from your own messages, never from the assistant's replies or from documents and web pages it reads. After a conversation, it looks for durable facts worth keeping, such as:
- Facts: "Works on credit-risk models", "Company fiscal year ends in March"
- Preferences: "Prefers concise answers", "Wants code examples in TypeScript"
Passing details and one-off questions aren't stored. When a fact changes, the newer one replaces the old one.
How it shows up in chat
Recall is mostly silent: when you start a chat, the assistant already has the facts relevant to it and uses them. When it looks something up specifically, you'll see a Recalling memory step in the conversation. To check what it knows, ask in any chat: "What do you know about me?"
Managing your memories
Open Settings → Memory to see everything stored, under Stored Memories. Each entry shows the fact, its type, and a link to the conversation it came from, with the date it was saved.
To remove one memory, click the trash icon on its row. It's deleted at once.
To keep a memory exactly as written, pin it with the pin icon on its row. Automatic upkeep leaves a pinned memory alone: Dream Mode never merges, refiles or retires it, and it's never dropped to make room when your memory fills up. Only you can edit or unpin it. The filter above the list shows only pinned or only unpinned memories.
Dream Mode: nightly consolidation
Over weeks of chats, your memory collects near-duplicates and facts that have gone out of date. Dream Mode cleans them up once a night. It re-reads your stored memories and:
- merges duplicates
- retires facts that have gone stale
- refiles memories kept under the wrong type
- picks up durable new facts worth keeping
Dream Mode is off by default and has its own switch. With memory on, turn on Enable Dream Mode at Settings → Memory. From then on it runs by itself. To start a run yourself, click Re-run now. If a run is already in progress or waiting for review, it asks before discarding that run.
Your chats keep using your current memories until a run is finished and applied. What happens to a finished run depends on your organisation:
- You review first (default). The run's changes wait under Pending changes. Approve or reject them one at a time, or use Accept all / Reject all. A rejected change won't be proposed again. If you edit your memories while a run waits for review, its proposals expire, so click Re-run now to get fresh ones. Each change is one of these:
- a merge, which folds duplicates together
- a refile, which moves a memory to a different type
- a forget, which retires a stale or contradicted memory and says why
- a plain edit, shown as a before-and-after
- Applied automatically. If your administrator has turned on auto-apply, a finished run updates your memory on its own. Last dream changes shows what it did, read-only.
If a run has nothing to change, you'll see a No changes card. If a run can't finish, your memories stay as they were. Once Dream has a few runs behind it, Consolidation activity charts what each night cost.
Dream Mode reads and rewrites only your own memories, within the same privacy and tenant boundaries as memory itself, and nothing leaves your environment.
Asking the assistant to forget
You don't have to open settings. You can ask in any chat:
- "Forget that I work at Acme": the assistant permanently deletes the stored memories that closely match what you describe, and tells you which facts it forgot. The match is strict: if nothing matches closely enough, it deletes nothing and says so. Be specific. If a related memory survives, ask again with more precise wording.
- "Delete all my memories": the assistant asks you to confirm in the conversation first. Once you clearly confirm, it wipes everything stored and reports how many memories were deleted.
These show up as steps in the chat: Forgot: … for a specific request, or Forgot all memories for a full wipe.
If you've turned memory off, the assistant can't delete memories for you. Remove anything still stored at Settings → Memory.
Deleting everything
Delete All, in the Danger Zone of Settings → Memory, permanently removes every stored memory after you confirm. You can also do this from chat.
Deletion is permanent. The fact is erased, and the assistant won't have it in future chats.
Turning memory off
Turn off Enable memory at Settings → Memory, and PrivateMind stops forming new memories. Anything already stored stays until you delete it, so switch off and Delete All if you want a clean slate.
Where next
- Chat: the conversation surface where memory is used.
- Conversation history: find, rename, and manage past chats.
- Account settings: the rest of your personal settings.