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

# Search Scanned PDFs and Images with On-Device OCR

> Elephas reads the text in your scanned PDFs, screenshots and photos of pages with OCR that runs on your computer. How it works, its limits, and scan tips.

Elephas reads the text in your scanned PDFs and images with OCR (optical character recognition) that runs on your computer. Add a scan or a screenshot like any other file, and you can ask about it and click through to the exact page.

## Before you start

* **Text recognition must be installed.** It downloads with Elephas's other components on first run, as **Text recognition for scanned pages**. Check it in **Settings › Downloads**. See [manage downloads](/docs/settings/manage-downloads).
* **English text is supported.** Text recognition reads English. Scans in other languages may come out garbled or empty.
* Supported scans: PDF, PNG, JPEG and TIFF. See [supported file types](/docs/brains-and-documents/supported-file-types).

## Steps

<Steps>
  <Step title="Check text recognition is installed">
    Open **Settings › Downloads**. **Text recognition for scanned pages** should say **Downloaded**. If it says **Missing**, click **Download** next to it.
  </Step>

  <Step title="Add your scans">
    Open a Brain and drag your scanned PDFs or images onto the page, or click **Add › Files…**. See [add files and folders](/docs/brains-and-documents/add-files-and-folders).
  </Step>

  <Step title="Wait for them to be ready">
    Scans take longer than ordinary files, because each page is turned into an image and read. The row shows **Extracting…** while this happens.
  </Step>

  <Step title="Ask and check the page">
    Ask your question. Citations to a scan point to its page. Click one to see the passage.
  </Step>
</Steps>

## How it works

### OCR runs on your computer

Text recognition is a small component that runs on your Mac or PC. Your scans are never uploaded to be read. Indexing happens on your computer. Always. After that, asking about a scan works like any other document: see [what data is sent to AI](/docs/privacy-and-security/what-data-is-sent-to-ai).

### What gets read, and when

| What you add | What happens |
| - | - |
| **Image** (PNG, JPEG, TIFF) | Read with OCR when you add it. Each frame of a multi-page TIFF is a page. |
| **Fully scanned PDF** (no text on any page) | Read with OCR when you add it. |
| **Mixed PDF** (some pages have text, some are scans) | The text pages are ready first. The scanned pages are read afterwards in the background, and the row shows **Recognising text…** while that runs. The document is searchable the whole time. |

Elephas reads up to the first 500 scanned pages of a document. Pages after that are not read.

Lines that OCR is not confident about are left out rather than guessed. If some pages had no readable text, the row says so quietly, for example **3 pages have no readable text.** The rest of the document still works.

### Limits

* Images up to 50 MB and 50 million pixels.
* Up to 500 scanned pages per document.
* Text in pictures on PowerPoint slides is not read yet.

## Tips for better results

* **Scan at 300 DPI or more.** Elephas reads pages at 300 DPI. A low-resolution scan loses detail that cannot be recovered.
* **Keep pages straight and flat.** Skewed, curled or rotated pages read poorly. Use your scanner app's auto-straighten if it has one.
* **Black text on a plain background works best.** Avoid colored paper, heavy shadows and photos taken at an angle.
* **Crop to the page.** Remove dark borders and desk edges.
* **Prefer the original file.** If you have the digital Word or PDF original, add that instead. Its text is exact.
* **Check a citation.** After you ask, open a cited page to confirm the text was read correctly before you rely on it.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Text recognition is not installed, so this image could not be read">
    The text recognition component is missing or damaged. Open **Settings › Downloads**, click **Download** next to **Text recognition for scanned pages**, then click **Retry** on the file. See [manage downloads](/docs/settings/manage-downloads).
  </Accordion>

  <Accordion title="No readable text was found in this file">
    OCR could not find any text it was confident about. The scan may be too faint, too small or not in English. Rescan at a higher resolution, then remove the file and add it again. See [file import errors](/docs/troubleshooting/file-import-errors).
  </Accordion>

  <Accordion title="Answers miss things I can see in the scan">
    Check whether the row says some pages have no readable text, and whether the document is longer than 500 scanned pages. Messy handwriting, stamps and tables in scans are often not read reliably.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="Supported file types" href="/docs/brains-and-documents/supported-file-types" />

  <Card title="Manage downloads" href="/docs/settings/manage-downloads" />

  <Card title="Add files and folders" href="/docs/brains-and-documents/add-files-and-folders" />

  <Card title="Citations and sources" href="/docs/ask/citations-and-sources" />
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.