Skip to main content

Updated 2026-11-20 View as Markdown

Bring an Obsidian vault into memory

Copy an Obsidian vault's notes into a workspace, folders and daily notes included, and know before starting what stays behind in Obsidian.

Works in
  • Hosted workspace
Used from
  • Its own AI
  • ChatGPT, Claude, Codex or Cursor
Not for a self-hosted install

An Obsidian vault is a folder of Markdown notes, so a workspace can take a copy of it with no export step. Each note becomes a page in Memory, the vault’s folders become folders there, and the workspace answers from the notes with the note it used.

The copy is a copy, not a connection. Nothing in the vault changes, and a note written in Obsidian afterwards stays in Obsidian until it is brought in too.

On the left, an Obsidian vault: a folder holding the folders Daily, People and Projects, the note Welcome.md, and the .obsidian settings folder. Arrows lead to two ways in: drop a zip in a chat, up to 100 MB from any browser; or Claude Code reads it, any size, on the AI's own plan. Both lead to Memory, where the vault is one folder under Sources, Harbor Notes, holding Daily, People, Projects and Welcome, without the settings folder.
Two ways in, one place it lands. A vault up to 100 MB goes through a chat. A larger one, or one that keeps changing, is read by Claude Code or Codex on the computer that holds it.

What comes across

  • Every note, whole, as one page in Memory under Sources, in a folder named after the vault.
  • The vault’s folders, as folders inside it. Daily notes come across like any other note, so a question about a day is answered from that day’s note.
  • [[Links]] between notes, as links on the page, under the words the note gave them. The workspace’s AI reads them when it answers. Opening one from a note in Memory is the exception, below in Links between notes.
  • Properties, the block of fields at the top of a note, as text at the top of its page under “Original note properties”. Memory keeps their words, not its own tags or details for them.
  • #tags, as words in the note. Search finds them; Memory does not list the note under that tag.
  • Tasks, as checkboxes, ticked or not as they were.
  • A canvas, as its text: the words on its cards and the notes it names, not the drawing.

Moving it in through a chat

A chat takes a file up to 100 MB. Notes are small, so most of a vault’s size is its pictures and attachments, which do not come across anyway. A vault over 100 MB fits once it is zipped without its attachments folder.

  1. Zip the vault folder. On a Mac, right-click the folder in Finder and choose Compress. On Windows, right-click it in File Explorer and choose Send to, then Compressed (zipped) folder.
  2. Drop the zip into a chat in the workspace and say what it is: “This is my Obsidian vault, zipped. Bring it into memory.”
  3. Read the reply. The workspace unpacks the zip, files each note, and says what it filed, what it skipped and why. A small vault is in within a couple of minutes.
A chat titled This is my Obsidian vault, zipped. The file harbor-notes.zip sits above the message This is my Obsidian vault, zipped. Bring it into memory. The workspace answers that the vault is in memory, 5 of 8 files filed as the collection sources/harbor-notes: welcome, projects/pier-rebuild, people/ada-park, daily/2026-11-18 and board, the canvas. It says each page keeps the note's full text with its wikilinks and tags as written, then lists what it skipped: a PDF with no text and an image.
A vault of five notes, dropped into a chat. The reply names every note it filed and every file it left out, with the reason.

A zip larger than 100 MB, up to 2 GB, goes through Google Drive instead, with no terminal: put it in Drive, give the workspace Drive at Read on Connectors, and ask in a chat to fetch it by name and bring it in. Bring a folder of documents into memory has the steps.

Moving it in with Claude Code or Codex

Claude Code and Codex read a folder on the computer they run on, with no size limit, on their own plan rather than the workspace’s credit. This is the way for a very large vault, and for a vault that keeps growing: run it again later, and it files only the notes that are new or changed.

Follow Bring a folder of documents into memory: connect the AI to the workspace with Read & Write, and paste the prompt there with the vault’s folder as the folder and a short name for the collection, like harbor-notes. Add one line to the end of the prompt, “Skip the .obsidian folder”, so the vault’s settings are not read as notes. The notes arrive under Memory, then Sources, then the collection’s name, the same place a chat puts them.

Check it worked

Open Memory, then Sources, then the vault’s name. The vault’s folders are there, and each note opens to its text.

Memory open on a workspace, in Sources, then Harbor Notes. Inside are the folders Daily, People and Projects, each with one item, and the pages Board and Welcome.
The vault, as it was. Its folders under Sources, and its notes inside them.
The Welcome note in Memory. At the top, the note's original properties as a block of text: tags harbor and onboarding, aliases Start, status active. Then the note: Welcome to Harbor Notes, a line with a code word, and Start with Pier Rebuild and talk to Ada, who runs it, where Pier Rebuild and Ada are links. The embedded picture shows as its text, pier-sketch.png in double brackets.
One note, with its links. The properties sit at the top as text, and the embedded picture stays a line of text.

Then ask the workspace about something only the vault would know, like a line from last week’s daily note. It answers from the notes and names the one it used.

Links between notes

A link comes across with its words, and Memory draws it as a link. Obsidian names the note a link leads to by its title, [[Pier Rebuild]], and Memory files that note under an address of its own inside the vault’s folder. So opening a link from a note in Memory today leads to a page that says it could not be read, where it should lead to the note. Open a linked note from its folder instead, or ask the workspace about it: its AI finds the note by the link’s words.

What stays behind

  • Pictures and attachments. Memory holds text, so an image, a PDF or any other file in the vault is not copied. An embed, ![[sketch.png]], stays in the note as that line of text. A PDF with text in it can be dropped into a chat on its own, up to ten at a time.
  • Embedded notes. ![[Another note]] arrives as that line, not as the other note’s text. The other note comes across on its own page.
  • Callouts, as quotes, with Obsidian’s marker such as [!note] left in as text.
  • What a plugin works out. A Dataview table, a Tasks query or a chart is drawn inside Obsidian, so the note keeps the query as written, not its result.
  • The .obsidian folder: the vault’s settings, theme and plugins. It is left out.
  • A canvas’s layout. Its cards’ words come across; where they sit and how they are joined does not.

The copy in memory does not follow the vault on its own. When the vault has moved on and the difference matters, bring it in again with Claude Code or Codex, which files only what is new or changed.