Importing from Logseq

Bring a Logseq graph into Mimir: hand this page to an AI assistant and it writes and runs the import, journal days and links included.

This page is written for you and for an AI assistant at the same time. You give it to your assistant, and the assistant writes a script that sends your Logseq pages into Mimir, one page at a time. Your journal days land on the same days in Mimir’s journal, and your links keep working.

Ask an AI assistant to import your Logseq graph.

Before you start

The import reads a folder of Markdown files:

Logseq OG keeps your graph as Markdown already. Use the graph folder, the one with journals/ and pages/ in it. Org-mode (.org) pages aren’t covered.
Logseq 2.0 keeps your graph in a database. Export it first: open the menu with three dots, choose Export graph, then Export as standard Markdown (no block properties). Use the folder it produces.

Images and other files in assets/ stay behind. A bullet that showed an image keeps the image’s name as text.

Then work through it in this order:

Back up your Mimir notes. See Backing up and restoring.
Import five pages, including one journal day.
Check how those pages look in Mimir.
Import the rest.

Running the import twice adds every page a second time. If the test pages look wrong, restore your backup before you try again.

For the assistant: the route

Mimir has no import command and no file format to write. Don’t write into Mimir’s database file. It holds a history of changes, not a table of bullets. Send each Logseq page through one link:

mimir://append?page=<title>&text=<text>

<title> and <text> are URL-encoded, with spaces as %20, not +. Mimir creates the page if it doesn’t exist and adds the text to the end of it. Page titles match without regard to case or repeated spaces, so Book Notes and book notes are the same page. On macOS, open each link with open:

open "mimir://append?page=Groceries&text=..."

Send one link at a time, with a pause of a second or two between them. Mimir opens each page as its text arrives.

A long page may not fit in a URL. For those, put the page’s text on the clipboard (pbcopy on macOS) and open the link without text: mimir://append?page=<title>. Mimir pastes the clipboard instead. Send each page whole, in one link. Each link adds to the end of the page, so a page split across links has to go in order.

For the assistant: page titles

Journal days. A file in journals/ becomes the page YYYY-MM-DD. The default file name is 2026_10_11.md, which becomes 2026-10-11. Mimir treats any page titled like that as that day’s journal entry, so append with that title lands on the right day. If a day already has bullets in Mimir, the imported bullets go below them.
Other pages. Use the page’s title:: property when the file has one. Otherwise use the file name without .md, decoded: %2F and ___ are /, and other %XX codes decode as usual. Namespaced titles such as Projects/Mimir stay as they are, slash included.
Formats in logseq/config.edn. :journal/file-name-format sets how journal files are named, and :journal/page-title-format sets how links to journal days are written inside bullets. Read both before you convert, and fall back to the defaults (yyyy_MM_dd and MMM do, yyyy) only when they’re missing.

For the assistant: how the text becomes bullets

Mimir splits the text you send into bullets, one per list item. Logseq’s own file shape already fits: each - line becomes a bullet, tabs or spaces in front of it become nesting, and an indented line without a dash joins the bullet above it as a second line. Send the page body as Logseq wrote it, after the conversions below.

Look at a few files before you write the script. If the bullets in them aren’t - lists, which can happen with a Logseq 2.0 export, put - in front of each paragraph, and check whether logseq/config.edn exists before relying on it.

Watch for these:

A page with one bullet arrives as a single bullet with the text exactly as sent, so strip its leading - first.
Lines before the first - (page properties, usually) make Mimir nest the whole page under them. Turn each such line into its own - bullet at the top of the page.
Code blocks inside a bullet. Lines in a fenced block that start with - or * split off as separate bullets, and indentation inside the block flattens. Check pages with code by hand after the import.
Front matter, the block between --- lines at the top of a file, becomes plain text. Turn its lines into - key: value bullets, the same as page properties.

For the assistant: converting Logseq syntax

Convert each bullet’s text before sending it. Leave anything inside `code` or a code fence exactly as written. When this table says to keep something as text, keep the words and drop only what the table names.

Logseq	In Mimir
`[[Page]]`	Keep. Same syntax, a link to the page.
`#tag`, `#[[two words]]`	`[[tag]]`, `[[two words]]`. Mimir has no tags. Only a `#` at the start of a word counts, so leave the `#` in a URL alone.
`[[Oct 11th, 2026]]` (a link to a journal day)	`[[2026-10-11]]`. Use the title format from `config.edn`.
`[label]([[Page]])`	`[[Page]]`
`[label](https://…)`	`label https://…`. Mimir has no Markdown links and shows bare URLs as links.
`![name](../assets/…)`	The file name as text, for example `photo.png`.
`TODO`, `DONE` at the start of a bullet	Keep. Mimir reads `todo` and `done` there as its to-do marker, in any capitalization.
`LATER`, `NOW`, `DOING`, `WAITING`, `IN-PROGRESS`	`TODO`
`CANCELED`, `CANCELLED`	`DONE`
`[#A]`, `[#B]`, `[#C]`	Keep as text.
`SCHEDULED: <…>`, `DEADLINE: <…>`	Keep as text, on the bullet’s second line.
`:LOGBOOK:` … `:END:`	Remove, lines and all.
`key:: value`	`key: value`, as text on the bullet’s own lines.
`id::`, `collapsed::`, `heading::`	Remove.
`((uuid))`	The text of the bullet with that `id::`, then a link to its page: `the bullet's text ([[Source page]])`.
`{{embed ((uuid))}}`	Same as `((uuid))`. Its child bullets don’t come along.
`{{embed [[Page]]}}`	`[[Page]]`
`{{query …}}`	Keep as code: `{{query …}}`. Mimir has no queries.
`{{youtube https://…}}`, `{{video https://…}}`	The bare URL, alone in its bullet. Mimir offers to embed a video link on its own line.
`==highlight==`, `^^highlight^^`	Remove the markers, keep the words.
`~~strikethrough~~`	Remove the markers, keep the words. Mimir hides text between single tildes, so `~~` would hide it.
`bold`, `italic`, `code`, code fences	Keep. Same syntax.
`# Heading` at the start of a bullet	Keep. Same syntax.
`> quote`	Keep. Mimir shows it as a guest block, someone else’s words.
`#+BEGIN_QUOTE` … `#+END_QUOTE`	`>` in front of each line inside. Drop the two marker lines.

To resolve ((uuid)), first read every file and collect each id:: with its bullet’s first line and page title. Use that first line only, converted, with any TODO-style marker removed. Its other lines and child bullets stay behind. A reference whose id:: isn’t found stays as ((uuid)) text.

For the assistant: checking the test pages

After the five test pages, ask the person to open them in Mimir and check:

The journal day shows up on the right date in the journal.
Nesting matches Logseq.
Links go to the right pages, and no page has a title like Oct 11th, 2026.
No text is masked. Masked text points to tildes left in the text.

Only then import the rest.