> ## Documentation Index
> Fetch the complete documentation index at: https://docs.oncortex.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Uploading files

> Drag in a document and let Cortex read it

Uploading is how you add a document to a brain directly, without going through an agent. Open the brain's **Inbox** and drag a file in, or use the upload button. It arrives as a new inbox item with its text extracted, ready to be filed.

<Tip>
  If you have an agent connected, the easiest way to add something is usually just to [ask it](/capture/asking-your-agent). Uploading is best for adding files, or for when you have not connected an agent.
</Tip>

## What you can upload

| Format                  | What happens                                                |
| ----------------------- | ----------------------------------------------------------- |
| Markdown and plain text | Used as-is.                                                 |
| PDF                     | The text is extracted, including from multi-page documents. |
| Word (`.docx`)          | The document's text is extracted.                           |
| PowerPoint (`.pptx`)    | The text from every slide is extracted.                     |

Each file can be up to 50 MB. Whatever you upload, Cortex pulls out the words so the content becomes searchable, even a slide deck becomes text your agent can recall.

## Say where it lives

If the file you are uploading lives somewhere, a SharePoint library, OneDrive, Google Drive, a Confluence page, paste that link into **Link to the original** in the upload window, and the version if you know it. Cortex keeps the copy you uploaded as the record of what was read, and keeps the link beside it, so a page that cites the document can also point at the live copy.

The link does one more thing: when you upload a newer version of the same document later, Cortex recognises it as a revision of the earlier one, links the two, and shows what changed between them. See [revisions](/capture/the-inbox#a-newer-version-of-a-document) on the inbox page.

## Say what it is

The upload window has a **File as** choice: Auto, General, Meeting transcript, Document or Agent conversation. Leave it on Auto and the filing pass first asks a cheap check whether the item is worth filing and which kind it is. Pick a kind and you have answered both questions yourself: the check is skipped, nothing is spent on it, and the item goes straight to that filing skill. You uploaded it on purpose, after all. The item's log records "skill pinned by the caller" so you can see which route it took.

## What extraction does, and does not, do

Extraction reads the words. It does not try to interpret images, charts, or the visual layout of a page. If a crucial fact only exists as a picture, it will not be captured. For anything text-based, extraction is reliable.

## Duplicates

Upload the same file twice and Cortex recognises it rather than making a second copy. It points you back to the item you already have. This means you can re-upload without worrying about clutter. [More on how duplicates are handled.](/capture/the-inbox#duplicates-take-care-of-themselves)

## Try it

1. Open a brain's Inbox and drag in a PDF.
2. Open the new item and read the extracted text.
3. Drag the same PDF in again and watch Cortex point you back to the first item.