How Memories work

Memories is your private, encrypted space for viewing, editing, importing, and exporting your complete AI memory . Your memory is encrypted with AES-GCM on your device — fully portable and fully yours — and it follows you across every model.

What makes Memories different

Data Ownership — Your memory belongs to you. Anuma never trains on it, sells it, or uses it without your consent.
Memory Management — Import, export, upload, review, edit, search, delete, or set your memories to private anytime.
Private by default — Memory is encrypted at rest and in transit. Only you and the AI tools you authorize can access it.
Multi-model Access — One unified memory across Claude, ChatGPT, Gemini, and more. Your preferences and context follow you.
Export anytime — Download your full memory anytime. No lock-in, no hidden dependencies.
Always up to date — Anuma continuously learns from your interactions and keeps your memory current without manual effort.

Your memories as a graph (web)

On the web, the Memories page is a Memory Graph: every memory Anuma saves, linked to the people, places, and topics it mentions.

Two views. On desktop, a Graph / List toggle switches between the graph canvas and a list view — a table where topics act like folders for bulk edits and privacy changes. On smaller screens you get the list view.
Reading the graph. Nodes are memories (brighter = more recent) and topics; connections link topics to their memories, similar memories to each other, and memories updated in the current session. Rings mark new, forgotten, and private memories. Open the Legend for the full key.
Click any node to open it — edit the text, add or remove topics, change its privacy, or delete it. Click a topic node to see every memory linked to it.
Live updates. While you chat, Just remembered and Cited chips show memories being saved and drawn on in real time.

Steps

Step 1: Open Memories

Tap Memories in the sidebar.

Step 2: Search and filter

The search bar finds memories and topics by meaning, not just keywords — it uses the same on-device semantic search as chat, so “the hotel I liked in Tokyo” finds the right memory even without the exact words. The filter narrows by privacy (All / Public / Private) or by topic, and can include forgotten memories.

Step 3: Organize with topics

Every memory carries topic chips that Anuma generates automatically (and you can edit). In List view, the topics row plays the role folders used to — select a topic to review or bulk-edit everything under it. Folders you created earlier were converted into topics.

Step 4: Edit or delete a memory

Open any memory to edit its text, topics, or privacy, or delete it. In List view, use the checkboxes to select multiple memories for bulk actions.

Step 5: Import and export

The Import & export menu lets you import conversations from ChatGPT, Claude, or Grok, and export your memories as a portable file. See Exporting your memory.

Choosing what stays private

Every memory is either Private or Public, and two controls decide which:

The privacy eye — the eye icon in the chat header, on web and mobile. With the eye open (Public mode), new memories saved from that chat are Public; with the eye slashed (Private mode), they’re saved Private. New accounts start in Private mode, and the eye remembers your choice on each device.
Per-memory control — every memory has a Keep this private toggle when you add or edit it, and the List view offers Make private / Change to public, including in bulk.

Private memories never appear anywhere public. Like every memory, they’re stored encrypted on your device and travel only inside your end-to-end-encrypted backup and sync — they’re never published or shared. Public memories can help shape what you choose to share through Anuma’s social features.

Two things the privacy eye does not do:

It doesn’t stop remembering. Private mode still saves memories — privately. If you want a chat that saves no memories at all, use a Temporary chat.
It doesn’t limit which models see your memories. Whatever a memory’s privacy setting, every model you chat with can draw on your relevant memories to help you. To control which model families you use at all, set Model privacy instead.

How Anuma keeps your memories tidy

Anuma does some housekeeping automatically, on web, iOS, and Android alike:

Automatic topic labels — Memories you add by hand or import get AI-generated topic chips within a few seconds; a brief Tagging memories… hint shows while this runs. Topics you’ve edited yourself are never overwritten.
Junk filtering — Auto-saved and imported memories are screened so obvious junk never lands in your vault: empty fragments, verbatim echoes of a chat turn (like a memory that’s just Said: 'tiger' from a word game), or a “fact” that’s simply your own name. A one-time cleanup also removed junk that was saved before this filter existed. Removed junk doesn’t appear in the Archived section and there’s no restore option for it — but the filter is deliberately conservative to make mistakes unlikely: it only matches these strict junk patterns, memories you typed in yourself are never touched, and the cleanup cancels itself rather than run if it would remove a large share of your vault.

Archived memories

Memories that age out of active recall move to a collapsible Archived section instead of disappearing. Expand it to review them — each one has a Restore button that returns it to your active memories. Archived memories you don’t restore are permanently deleted 30 days after archiving.

Held for review

If incoming content looks unsafe to store — for example, text that resembles a prompt-injection attempt — Anuma quarantines it instead of saving it. The vault shows a Held for review count so you know it happened; the quarantined content is kept out of recall and is never shown or used. If the count is zero, the notice doesn’t appear at all.

On the mobile app

The iOS and Android app keeps a list-style Memories view with the same content: semantic search, typed memory badges, the Archived section, per-memory privacy, and Export memories (a readable Markdown copy via the share sheet). The privacy eye sits in the chat header there too.

Memories on iOS

Good to know

Your memory is encrypted with AES-GCM and stored on your device. Anuma cannot access it.
Memory works across all models. Switch from ChatGPT to Gemini, and your context stays with you.
Memory syncs across web, iOS, and Android. Switch devices and your AI knows the same things everywhere.
You can import memories from other AI providers to give Anuma a head start, and on web you can import them from uploaded documents (.pdf or .docx, with OCR for scanned files).
In any conversation, open the Memories panel (web) or sheet (mobile) to see exactly which stored memories that chat is drawing on. The list builds up as the conversation grows and persists across reloads.

Related articles