Document Parser
The Document Parser component — read a document's text from PDF, Word, Excel, or text files, using OCR for scanned pages.
The Document Parser reads a document and returns its text. Word headings and tables, Excel sheets, and pages read with OCR come back as Markdown, keeping their structure; a PDF's text layer comes back as plain text. Give it a file and hand the text to an agent, a Prompt, or an Embeddings step. It's the same reader datasets use.
What it reads
| File | How it's read |
|---|---|
| Page by page from the text layer. Pages that are only images — scans — need OCR. | |
Word (.docx) | Headings and tables are kept. |
Excel (.xlsx) | Each sheet becomes a heading and a table, with calculated values. |
| Text | Markdown, CSV, and plain text are read as they are. |
| Images | Only with an OCR connection. |
Anything else fails the step, naming the file.
Inputs
- document — the file to read, bound from the Input or an earlier step. It can also be a list of files — see below.
- OCR — optional. Pick a connection with the OCR capability, such as Azure Document Intelligence, to read scanned pages.
Settings
| Setting | Default | What it does |
|---|---|---|
| Always use OCR | Off | Read every page with OCR, not just the ones with no text layer. Better tables and reading order — and a charge for every page. |
| Allow image-only pages | Off | When a PDF has scanned pages and there's no OCR connection, return what text there is instead of failing. |
| Parallel files | 5 | How many files to read at once, when the input is a list. |
Scanned documents
With an OCR connection, only the pages that need it go to OCR (unless Always use OCR is on). Without one, a document with scanned pages fails, and the error names the pages — "Pages 3, 5 of 'statement.pdf' are images, not text." That's on purpose: a partly-read document is usually worse than an error, because nothing downstream can tell what's missing. Turn on Allow image-only pages only where that gap is acceptable.
Reading many files
Bind document to a list of files — the pages from a Document Splitter, say — and the parser reads every one, Parallel files at a time, and returns a list of their text in the same order.
Input (file) → Document Splitter → Document Parser → Agent → OutputAs a tool
The Document Parser can be an agent's tool, so the agent can read a file when it decides to. As a tool it has no OCR connection, so scanned pages fail — use it as a step when you need OCR.