Claude chats are often long and full of things worth keeping: scripts, drafts, plans written as artifacts. Copying them by hand loses most of that. This guide shows how to get them into Obsidian as real notes.
What Claude chats need
- Artifacts. Code and documents Claude writes in the side panel are not part of the visible message text. A useful note has to include them.
- Thinking. With extended thinking on, the reasoning is worth keeping, but too long to read every time.
- Projects. Claude Projects group related chats; your vault should too.
- Updates. You come back to a chat days later. The note should follow.
Set it up
- Add Chat to Markdown to Chrome or Edge.
- Click Choose a folder on the welcome page and select a folder in your vault for the chats, such as a new
AI Chatsfolder. Choose Allow on every visit when Chrome asks. - Turn on Claude on the same page (Chrome asks to allow claude.ai) and stay signed in to claude.ai in this browser.
- Press Start saving. Claude chats go to
AI Chats/Claude/.
New and changed chats are saved every 3 hours by default, or on any schedule you set, down to every 15 minutes. To save a chat right now, click Save next to Share at the top of the chat on claude.ai. Its menu has Copy as Markdown, for pasting a chat somewhere else.
Artifacts, thinking and tools
A saved Claude note looks like this in source mode:
## You
Write a script that renames my photos by the date they were taken.
## Claude
> [!abstract]- Thinking
> The user wants the EXIF capture date, not the file date…
This reads each photo’s EXIF date and renames it.
**Artifact: rename_photos.py**
[[rename_photos 5c0e9a1d.py]]- Artifacts are on by default. Each artifact, Claude Docs document or file Claude writes is saved as its own file in
attachments/next to the note and linked where it appeared. When Claude changes it later in the chat, the file is rewritten with the latest version. Prefer one self-contained note? Turn off Artifacts as files, and they are written inline as fenced blocks in the right language. - Thinking and tool calls (such as web searches) are off by default. Turn them on in Settings → What to save, under Model thinking and Tool calls. Thinking is saved as a folded callout, which other Markdown apps show as a plain quote, or as plain text if you pick Plain under “How thinking looks”. A tool call is saved as its name and its input as JSON.
Claude Projects as folders
Chats that belong to a Claude Project are saved in a subfolder named after it, such as AI Chats/Claude/Thesis/, and the note gets a project property. If one project is private, add it to the skip list in Settings and none of its chats are written.
Chats you delete on claude.ai stay in your vault, marked deleted: true, so cleaning up your Claude sidebar never costs you a note. And when a chat changes, the version before is kept, so an edited prompt or a regenerated answer doesn’t erase what the note said.
One chat by hand
Claude has a copy button under each reply that copies it as Markdown. Paste the replies one by one into a note, and add your own messages in between. It works, but artifacts and thinking are left out, and the note won’t update. For everything at once, Claude’s data export is covered in how to export your Claude conversations.