> This page is for Document AI.

> Sarvam Voice Agents (Samvaad) is a voice AI agent platform for Indian languages: build, test and deploy AI calling agents on your own phone numbers, from ₹3 a minute. Start here: https://docs.sarvam.ai/voice-agents/why-voice-agents.
> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt. For a section index, append `/llms.txt` to any section URL (for example https://docs.sarvam.ai/voice-agents/llms.txt).
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Parse a document

> Turn a PDF or scan into clean text, tables, and layout, edit the detected sections, and download as Markdown, HTML, or JSON.

**Parse** reads every page of a document and turns it into structured text. Headings, paragraphs, tables, lists, images, and reading order are kept, so the output is ready for an LLM, a RAG pipeline, search, or an archive. Parse was called **Digitise** in earlier versions of the app.

It works on modern printed documents, scans, handwritten forms, land deeds, and historic multilingual manuscripts.

## Step 1: Add your document

Click **Parse** under **Products** in the sidebar (or the **Parse** card on Home). Drag and drop your files, click **Upload documents**, or click the **IT policy.pdf** demo document under **Try a demo doc**.

![Parse page with an upload area that accepts PDF, JPEG, PNG, TIFF, or ZIP, up to 10 files or 20 pages in total, an Upload documents button, and a Try a demo doc strip with IT policy.pdf.](/_fern-files/_img/d41f3c22d13575345b97bb2ba9c3a9e5d8e2b6ff2707b8b5dae1040a1c99d17e.webp)

> **Note**
>
> Accepted formats: **PDF, JPEG, PNG, TIFF, ZIP**. Each run takes up to **10 files or 20 pages in total**. Click **+ Add** under the page thumbnails to add more files to the same run, or the trash icon to remove one. For bigger jobs, use the [Parse API](/api/api-guides-tutorials/document-intelligence/how-to/parse).

## Step 2: Run Parse

Parse needs no setup. Click **Parse 1 document** (the label counts your files). The credits the run will use are shown under the button. A small single-page document typically finishes in a few seconds.

You can leave the page while a run is processing and reopen it later from **History**.

## Step 3: Review the output

The source document sits on the left with a **color-coded box** around every region Sarvam detected. The right pane has three tabs:

![Parse results for IT policy.pdf: the scanned page on the left with coloured boxes around the logo, title, table, and paragraphs; the Edit sections tab on the right lists numbered sections such as 1. Header, 2. Header, 3. Image Caption, and 4. Table.](/_fern-files/_img/b9ef2bd542ea187b15557bb1edb0f45095b364c362294708b120aad118c60f33.webp)

#### Edit sections

A numbered list of every detected region, in reading order. Each section has a **type** dropdown and its content underneath. Click into any text to correct it. Tables render as grids you can edit cell by cell. Images and signatures get a written description. Changes save automatically; the **Saved** badge confirms it.

#### Markdown

The document as Markdown, exactly as it will download. Headings, lists, and tables are already formatted.

#### Preview

A rendered, read-only view of the final document. To change the text, go back to **Edit sections**.

### Change a section's type

Open the type dropdown on any section to reclassify it. The list is searchable and includes:

| Group              | Section types                                                                         |
| ------------------ | ------------------------------------------------------------------------------------- |
| Titles             | Headline, Sub Headline, Chapter Title, Section Title                                  |
| Body text          | Paragraph, Ordered List, Unordered List, Quote, Formula, Sidebar                      |
| Page furniture     | Header, Footer, Footnote, Page Number, Folio                                          |
| Tables and visuals | Table, Table Caption, Image, Image Caption, Photograph, Chart, Diagram, Advertisement |

### Add a region Sarvam missed

If a caption, stamp, or column wasn't picked up, draw it in.

![Parse page in Edit layout mode, with a toolbar at the bottom of the source pane showing a confirm tick, a cancel cross, the hint Drag on the page to add a box, then Run again to read it, a Label dropdown set to paragraph, and a delete icon.](/_fern-files/_img/6f8b62ae541fe57a2b14c71fd57c43c38d736b888737cb3e36cb5daedbd82b31.webp)

#### Enter Edit layout

Click **+ Edit layout** in the toolbar under the source page.

#### Draw a box

Drag on the page around the region you want to capture.

#### Pick a label

Choose the section type from the **Label** dropdown. It uses the same list as the section types above.

#### Confirm and run again

Click the tick to keep the box, or the cross to discard it. Then click **Run again** so Sarvam reads the new region. Use the undo and redo arrows in the toolbar to step back through layout changes.

### Multi-page documents

The toolbar under the source page shows the page count (for example `1/3 pages`), and each page shows its own detected regions.

## Step 4: Download

Click **Download**, choose a **Format**, and click **Download** again.

#### Markdown

Clean, LLM-friendly text with heading levels and tables. Best for RAG, embeddings, and archives.

#### HTML

Rich layout with headings, tables, and image descriptions. Best for publishing or HTML-aware tools.

#### JSON

Structured output with each section and its type. Best when you need the layout as data.

## Run it from code

Click **View API code** to get cURL, Python, or JavaScript that runs this exact setup through the API. Replace `YOUR_API_KEY` and the file name, and you're ready to go. See the [Parse API guide](/api/api-guides-tutorials/document-intelligence/how-to/parse) for every option.

> **Tip**
>
> Parsed Markdown makes a much better LLM input than raw OCR. Headings give you natural chunk boundaries, tables stay as tables, and multi-column pages keep the right reading order. If your RAG pipeline ingests raw OCR today, swap in Parse output and measure retrieval again.

Next: [Extract fields](/doc-ai/how-to/extract-fields).