On this page · 6 sections #
Short answer
/mnt/data/pasted markdown.md
is a file path inside ChatGPT’s own temporary sandbox — a Linux container running on OpenAI’s servers. It is not a path on your computer, which is why clicking it does nothing. Use the download button next to the file in the chat instead. If the session has expired, ask ChatGPT to paste the content back into the conversation and save it yourself.
You asked ChatGPT to do something with a document. It replied with a confident-looking link — /mnt/data/pasted markdown.md
, or maybe file:///mnt/data/pasted markdown.md
— and clicking it does nothing at all. Your browser shrugs. Your file manager has never heard of it.
Nothing is broken. The path is just describing a place you don’t have access to.
What /mnt/data actually is #
When ChatGPT analyses a file, runs code, or generates a document, it does that work inside a sandboxed Linux container running on OpenAI’s infrastructure. /mnt/data
is the working directory inside that container — the place uploaded files land and generated files get written.
So /mnt/data/pasted markdown.md
is a real file. It’s just real over there, on a machine you will never touch, inside a container that is thrown away when your session ends.
The confusion is understandable: the model writes the path the way it would write any other file reference, because from its perspective that is where the file lives. It has no way to hand you a path on your own disk, because it doesn’t know anything about your disk.
Why the link fails, specifically #
When the path arrives as a link — often file:///mnt/data/pasted markdown.md
— your browser treats it as a local file URL and looks for /mnt/data/
on your machine.
- On
macOS, there’s no
/mnt
directory at all. Nothing happens. - On
Windows, the path isn’t even valid syntax for a local file. Nothing happens. - On
Linux,/mnt
exists but is almost certainly empty. Nothing happens.
Modern browsers also block file://
links embedded in web pages for security reasons, so even if the path did exist locally, the click would still be ignored.
You’ll see the same thing with numbered variants like /mnt/data/pasted markdown(1).md
— that’s just the sandbox avoiding a filename collision after you pasted content more than once.
How to actually get your file #
1. Use the download link, not the path. When ChatGPT generates a file, it normally renders a proper download element — a filename with a download icon — separately from any path it mentions in the prose. Click that, not the /mnt/data/...
text.
2. If the download link has expired, ask for the content directly. Sandbox sessions are temporary. Once one is cleaned up, the file is gone and no link will resurrect it. The reliable move is to ask:
“Output the full contents of that file directly in the chat, in a code block.”
Then copy it and save it yourself.
3. Save it with the right extension. Paste into a plain text editor and save as something.md
. On Mac, set TextEdit to plain text first (Format → Make Plain Text
). On Windows, pick “All Files” in Notepad’s save dialog so it doesn’t quietly append .txt
.
Don't retype it by hand
If the content is long, ask for it in a code block rather than as normal prose. Code blocks give you a copy button and preserve the exact markdown — including the leading #
symbols and indentation that formatting depends on.
Then: opening the .md file #
Once you have the file locally, you’ll hit the next small wall — double-click it and you get raw text full of #
symbols and asterisks, because neither macOS nor Windows renders Markdown out of the box.
A .md
file is a Markdown document: plain text with simple formatting conventions. To read it as intended you need a Markdown viewer. MDHero is free, open source, under 8MB, and can be set as the default app for .md
files so this stops being a recurring annoyance — on Mac and Windows.
If you’re doing this often, skip the file entirely #
Most people asking about /mnt/data
don’t actually want a file — they want the content, formatted properly. In that case there’s a shorter route: have ChatGPT print the markdown in the chat, copy it, and paste it straight into a viewer that renders markdown from the clipboard.
MDHero’s paste mode (Cmd+Shift+V
) does this in one step, and it also cleans up the escaped \n
characters that AI output tends to carry. No file, no download, no sandbox path. See copying ChatGPT output with formatting intact for the full workflow.
The short version #
/mnt/data/
is ChatGPT’s scratch space, not yours. The path is accurate from the model’s point of view and useless from yours. Download the file properly while the session is alive, or ask for the content in the chat and save it yourself — then open it in something that renders Markdown.
Frequently asked questions #
### What is /mnt/data? #
/mnt/data is a folder inside the temporary Linux sandbox that ChatGPT uses when it runs code or generates files. It exists on OpenAI's servers, not on your computer, and it is discarded when the session ends.
### Why doesn't the /mnt/data link open? #
The link points to a path on ChatGPT's server, not on your machine. Your browser resolves it against your own filesystem, finds nothing there, and fails. The file was never downloaded, so there is nothing local to open.
### How do I actually get the file? #
Use the download button ChatGPT shows next to the generated file rather than the inline path. If the session has expired, ask ChatGPT to output the content directly in the chat, then copy it and save it yourself as a .md file.
### How do I open the .md file once I have it? #
A .md file is a Markdown document. Any text editor will open it as raw text. To see it formatted, open it in a Markdown viewer such as MDHero, which is free, under 8MB, and available for Mac and Windows.