# Scanned PDFs & Images

> Make the text in scanned PDFs, photos and screenshots searchable — read on your computer, only for the files you choose.

Source: https://docs.docgraph.app/formats/scans/

A scanned contract, a photographed receipt or a screenshot has no text layer: search finds nothing in it, and the chat assistant knows nothing about it. DocGraph can read the text in those files **on your computer** and add it to search, so keyword search, the chat and *Ask about this file* reach it like any note.

Nothing is read unless you ask. A vault full of photos stays a vault full of photos.

## Which files are scans

A PDF whose pages have no text but do have a picture is a **scan**: its format badge in the file tree reads **Scan** instead of PDF. A page with no text and no picture is just a blank page, not a scan. Images keep their **IMG** badge.

Once a file is in search, its badge shows where it stands:

| Badge | Meaning |
|---|---|
| **Scan ✓** / **IMG ✓** | Its text is searchable. Hover for the page count, the reader that read it, the date, and any pages that were hard to read. |
| **Scan …** / **IMG …** | Waiting to be read, or being read now. |
| **Scan !** / **IMG !** | Something could not be read. Hover for the reason. |

## Make a file searchable

Right-click a scanned PDF or an image in the file tree and choose **Make Text Searchable**. The text is read in the background; the badge turns to ✓ when it is done.

- **Several files:** select them (⌘-click or ⇧-click) and choose **Make Text Searchable** from the selection's menu.
- **A whole folder:** right-click the folder and choose **Make Scans in Folder Searchable** — every scanned PDF and image below it.
- **A screenshot in a note:** click the image in the note and use **Make Text Searchable** in its toolbar. The screenshot's text is then found as part of *the note*, so a search for something in the screenshot opens the note.
- **From a chat answer or the viewer:** when a question touches a scan whose text is not searchable, the answer says "*report.pdf* is a scan" with **Make Searchable**, then **Ask again** once it has been read. Opening such a PDF shows the same line at the top of the page.

A file over 200 pages asks first and tells you how long it will take. A PDF whose text is already searchable is left alone.

### Remove a file from search

Right-click it and choose **Remove from Search**. Its text leaves search **and is deleted from your computer** — DocGraph keeps no copy. Making it searchable again reads it again.

### When the text looks wrong

Some scanners put their own, often poor, text layer inside a PDF. **Read Text Again** (right-click a file that is in search) throws away what was read and reads every page again, including pages that already had text.

## Read new scans automatically

For a folder a scanner app writes into — *Scans*, *Receipts* — right-click it and turn on **Make New Scans Searchable Automatically**. Scans and images added to it afterwards are read without asking; the ones already there are not (use *Make Scans in Folder Searchable* for those). **Remove from Search** on one of its files keeps that file out.

To read every scan in the vault, choose **Read all scans automatically** in Settings (below). Folders you list under *Skip folders* are never read automatically, and files over 200 pages wait for you to start them.

## Fast or Best quality

| | **Fast** | **Best quality** |
|---|---|---|
| Where | Built into macOS 13 or later | Runs on your Mac as a separate download |
| Download | None | About 1.7 GB, once |
| Speed | Instant | About 1–3 seconds a page, in the background |
| Reads well | Clean printed text, receipts, handwriting, non-Latin scripts | Tables, columns, forms, poor scans |

**Fast** works even with AI turned off: it uses the text reader built into macOS. **Best quality** uses a document-reading model that runs on your computer, separate from the chat model, so a long read never takes the chat's model away. On a computer with limited memory, it waits between chat turns.

Switching reader reads the files in search again with the new one; switching back reuses what the first one read, as long as it is still there.

## Settings → AI → Scanned documents

- **Read scanned documents** — on by default on macOS 13 or later.
- **Which files** — *Ask me per file* (the default) or *Read all scans automatically*.
- **Quality** — Fast or Best quality, with the model's download and **Remove model** (which also deletes the text it read).
- **Progress** — pages read so far, **Pause / Resume**, and why reading is waiting (paused, on battery, or waiting for the chat).
- **Pause on battery** — on by default.
- **Files in search** — every file whose text is searchable, each with a remove button, and **Read again** for all of them.
- **Skip folders** — folders the automatic modes never read; `*` and `**` match names.

The first time you set up AI, the setup asks about scans too — but only if your vault has scanned PDFs.

## Privacy

Everything is read on your computer; nothing is uploaded. The recognized text is kept outside your vault, in DocGraph's own data folder, like the search index, and **Remove from Search** deletes it. Encrypted files are never read. Citations in the chat say **recognized text** when a passage came from a scan, and **recognized text, low confidence** when the reader was unsure.

## Limits

- HEIC photos are read on a Mac with Fast only.
- Text inside Word or PowerPoint files' images is not read.