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

# Chat with files

> Chat directly with one or more files to get grounded answers with page/section citations. This is ideal for close reading, extracting methods/results, or comparing specific passages across papers.

### What "chat with files" means

When you start a chat from a file (or a selection of files), Clevia scopes its analysis to just those sources. Every answer includes citations that link back to the exact location in the file viewer so you can verify instantly.

<Tip>
  Use file-level chat when you know the few sources you want to analyze deeply
</Tip>

## Crafting effective prompts (for files)

Try these patterns:

* *Locate specific evidence:* "Extract all sample sizes and inclusion criteria from this paper. Cite each extraction."
* *Explain a passage*: "Explain the loss function described around pages 7–8 in plain English, and link equations."
* *Compare two methods*: "Compare the preprocessing pipeline between these two PDFs. Provide a 4-row table: step, paper A, paper B, notes."
* *Check claims*: "Does the conclusion restate results accurately? Quote and cite discrepancies."

## What to expect in answers

* Citations like \[1], \[2] show file short title + page/section
* Clicking a citation opens the viewer at the exact location

## Troubleshooting

<AccordionGroup>
  <Accordion title="No citations in the answer">
    * Enable Citations required and re-ask.
    * Confirm the files finished processing/OCR.
    * Narrow scope to the most relevant pages or a smaller file set
  </Accordion>

  <Accordion title="File not ready / not indexed">
    Wait until the file status shows Ready in the Library, then retry.
  </Accordion>

  <Accordion title="Slow or partial results">
    * Reduce the number of files or ask for specific sections (e.g., "Methods", "Table 2")
    * Split your question into two prompts (extract → synthesize)
  </Accordion>

  <Accordion title="Highlights don't match text">
    The PDF may be an image scan. See Import files → OCR support
  </Accordion>
</AccordionGroup>