Overview
With intelligent document processing turned on, AI agents in integrator.io read native and scanned PDFs, Word, Excel, PowerPoint, and RTF documents, TIFF and BMP scans, images, and plain-text files, and each file can be up to 20 MB. Celigo extracts the text and tables from each document — using optical character recognition (OCR) for scanned pages — before the AI model reads it, so the same file types work on every model, whether the agent uses a Celigo AI model or your own through a bring-your-own-key (BYOK) connection.
This article lists every supported file type, how each one is read, the limits that apply, and what happens when intelligent document processing is turned off.
Supported file types
| Document family | File types | How Celigo reads it | Notes |
|---|---|---|---|
.pdf |
Native text and tables from digital pages; OCR for scanned and image-only pages | Pages are read individually, so a PDF that mixes digital and scanned pages works. The first 50 pages are read. | |
| Word |
.docx, .doc
|
Text and tables; OCR for embedded images when the document has little text of its own | Some .doc files saved by Word for Mac can't be read; save them as .docx. |
| Excel |
.xlsx, .xls
|
Every sheet, as tables | Tell the agent which sheet to use when a workbook has several. |
| PowerPoint |
.pptx, .ppt
|
Slide text and tables | |
| Rich Text Format | .rtf |
Text | |
| TIFF |
.tif, .tiff
|
OCR on every page | Multi-page TIFF files are supported. |
| Bitmap | .bmp |
OCR | |
| Images |
.png, .jpg, .jpeg, .webp
|
The AI model's own vision | Images aren't processed by Celigo's OCR. To use OCR on a scanned page, save the scan as PDF or TIFF. |
| Plain text |
.txt, .csv, .json, .xml, .html, .htm, .md, .yml, .yaml, .log
|
Passed to the model as text | Plain-text files aren't processed; they reach the model as they are. |
Word, Excel, PowerPoint, and RTF files are read as one document, not page by page.
To process spreadsheet or CSV data one row at a time — with a separate agent call, error, and retry for each row — parse the file into records with a file definition first, instead of passing the whole file to the agent.
How Celigo identifies a file's type
Celigo decides how to read a file from the extension in its file name. If the file name has no extension, or an extension Celigo doesn't recognize, Celigo identifies the type from the file's contents — for example, a PDF, a legacy Word, Excel, or PowerPoint file, an RTF file, an image, or plain text. Map the file name, including its extension, to the agent's files input whenever the source provides it.
File types that aren't supported
A file type Celigo doesn't accept fails the record with "This file type is not supported. Supported types include images, PDF, DOC, DOCX, PPT, PPTX, XLS, XLSX." Examples:
| File type | What to do |
|---|---|
OpenDocument files (.odt, .ods, .odp) |
Save as .docx, .xlsx, .pptx, or PDF. |
Outlook message files (.msg) |
Pass the message's attachments to the agent, or save the message as PDF. |
Archives (.zip) |
Extract the files first, and pass each file to the agent. |
EPUB files (.epub) |
Save as PDF. |
Other image formats, such as GIF, reach the model as images, and whether the model can read them depends on the model.
Limits
| Limit | Value | What happens when it's exceeded |
|---|---|---|
| File size | 20 MB per file | The record fails with "This file exceeds the maximum supported size of 20 MB. Please upload a smaller file." |
| Total size of the files in one agent run | 50 MB | The run fails. See Input for Celigo AI agents. |
| PDF pages | The first 50 pages of each PDF | Pages after page 50 are ignored, without an error. |
| Processing time | Varies with the file | A document that takes too long to process fails with an error that begins "Document parsing timed out." Split large scanned files. |
| Amount of text | Depends on the model | A very large workbook, presentation, or document can produce more text than the model accepts in one request, and the run fails. Split the file, or choose a model with a larger context window. |
| Password protection | Not supported | The record fails with "Document is encrypted or password-protected." Remove the password, and retry. |
When one agent run includes several files, each file is processed, and the model receives all of them together. If any file can't be processed, the run fails, and the model doesn't receive a partial set of files.
Languages and content
- Languages — Text in digital documents, such as PDFs with selectable text and Office files, is read directly from the file, whatever its language. OCR, which reads scanned pages, recognizes languages written in the Latin alphabet, such as English, French, German, Spanish, Italian, Portuguese, and Dutch. It doesn't recognize other scripts, such as Chinese, Japanese, Korean, Arabic, or Russian. Test with your own documents first.
- Handwriting — Handwritten content isn't supported.
- Scan quality — Low-resolution, skewed, or faint scans reduce OCR accuracy. Rescan at a higher resolution when values are missing or wrong.
When intelligent document processing is turned off
With the setting off, each file goes to the AI model as it is, and the model's own file support decides what it can read: PDFs and images on models that accept them, plain text on any model, and no Word, Excel, PowerPoint, RTF, TIFF, or BMP files. A file the model can't read fails the record with "This model cannot read this file type. Please use a PDF, or convert the file to a supported format." See Turn on intelligent document processing.