MnemoverseDocs

Appearance

Export your memory ​

Beta

The console page described here is in beta: it is new and may fail. It only reads your memories, so nothing on it can change or lose them. If something looks wrong, tell us at support@mnemoverse.com. The REST path on this page is the same memory and works regardless.

Every memory in your account can be downloaded as a plain file whenever you want: from the console with one button per format, or from a script with one REST call. The console page puts it in its own words: "Download every memory in your account as a plain file, whenever you want. Your memories are yours."

In the console ​

Open Dashboard → Memory → Export in the console. Export and Import are two tabs under the same Memory entry; the import side is Bring your memory.

The page first counts what your account holds: current memories, replaced versions of corrected memories, how many Vault entries the Markdown, CSV and text files leave out, and any stored entry that could not be read (those are in the JSONL file exactly as stored and missing from the other three). Then it offers four downloads and one switch, Include replaced versions, which adds the earlier versions of corrected memories to the Markdown, CSV and text files. The JSONL file always has them.

Each file is named mnemoverse-memory-YYYY-MM-DD with its extension and is written as it streams, so nothing is held whole in the page. Once a download starts, the page warns: "A large account can take a few minutes."

The four formats ​

FormatFileWhat it holds
JSONL.jsonlThe complete copy: byte for byte what GET /memory/export streams. One memory per line with every field the export row carries, oldest first, replaced versions always included, Vault entries as their description only. Keep this one as your backup; it is also what Bring your memory reads back
Markdown.mdA readable list: a ## <domain> section per domain (general first, then the others sorted by name), and one - **YYYY-MM-DD** — memory (concepts: a, b) bullet per memory, dated by the day it was saved (created_at)
CSV.csvFor spreadsheets (RFC 4180): the columns created_at, domain, content, concepts (joined with ; ), external_ref and atom_id, one row per memory
Text for Claude or Gemini.txtOne [YYYY-MM-DD] - memory line per memory, oldest first, each memory flattened to a single line. The date is the memory's occurred_at when it has one, otherwise its created_at. See below

Markdown, CSV and text are views of the same stream: they never contain a Vault entry, and they leave out replaced versions unless the switch is on.

What is exported, and what is not ​

Exported:

  • Every memory you or your tools saved, with its date, domain and concepts. The JSONL file also carries each memory's ID, metadata, external_ref and the scores and counters listed under GET /memory/export.
  • Replaced versions of corrected memories: always in the JSONL file, where supersedes and superseded_by record which version replaced which; in the other three files only with Include replaced versions on.
  • Vault entries, in the JSONL file only, as rows with atom_type: "secret": their description as content and their secret:<alias> reference, never their value. The page counts them before you download anything.

Not exported, in the page's words: "How your memories connect to each other, and the search data Mnemoverse computes from each one. Those are built on top of your memories, not part of them." That means the links Mnemoverse learns between memories as you use them, and the embeddings computed from each memory's text. Importing the JSONL file again stores the memories; the links are learned again with use, as for any new memory.

Move your memories into Claude or Gemini ​

The text file is the format the memory-import prompts of other assistants ask for. Anthropic's recommended export prompt says "Format each entry as: [date saved, if available] - memory content" (Import and export your memory from Claude, read 2026-10-24). Each line of the file already has that shape:

text
[2025-03-14] - Prefers short answers with the code first
[2026-08-02] - Staging deploys need REDIS_URL set in the staging env group
  • Claude: open Claude's built-in import flow, paste the lines into its text box and click Add to memory. Anthropic's article says Claude "will extract key information and store it as individual memory entries", and that "Claude may not retain imported personal details unrelated to work."
  • Gemini: at gemini.google.com, open Settings & help → Import memory to Gemini. Google's steps have you run Gemini's prompt in the other platform, then "paste the copied response into the designated text field, and click Add memory" (Import from other AI platforms to Gemini Apps, read 2026-10-24). The text file stands in for that response, so paste its lines there. The same article says the feature needs a personal Google Account and an age of 18 or older, and that it "is currently not available in" the EEA, Switzerland or the United Kingdom.

Both assistants decide for themselves what to keep from what you paste. The file is plain text, so you can delete lines before pasting.

From a script (REST) ​

The JSONL download is the REST endpoint, so a script with an API key gets the same file:

bash
curl -H "X-Api-Key: mk_live_YOUR_KEY" \
  https://core.mnemoverse.com/api/v1/memory/export \
  -o mnemoverse-export.jsonl

The response is streamed JSONL; GET /memory/export describes each line. There is no format parameter on the API: Markdown, CSV and text are made by the console from this stream. To get the text format from a script, convert the JSONL yourself; this jq filter leaves out Vault entries and replaced versions and dates each line the way the console does:

bash
jq -r 'select(.atom_type != "secret" and .superseded_by == null)
  | "[\((.occurred_at // .created_at)[0:10])] - \(.content | gsub("\\s+"; " ") | ltrimstr(" ") | rtrimstr(" "))"' \
  mnemoverse-export.jsonl > mnemoverse-memory.txt

A paused account can still export: when a subscription's writes are paused or the account is past its memory limit, writes are refused while reads and export keep working (Subscription / atom limit).