Excel Support
Agentic Table Extractor supports Excel workbooks in .xlsx and .xls formats. Excel support extends the same 6-agent pipeline used for PDFs, with sheet-level processing.
How Excel Files Are Processed
Each sheet in an Excel workbook is treated as a separate "page" in the pipeline. The agents process sheets the same way they process PDF pages:
- Agent 1 (PresenceDetector) screens each sheet to determine if it contains the target table.
- Agent 2 (TableDetector) analyzes the table structure on relevant sheets.
- Agent 3 (ContentExtractor) extracts row data from each sheet and merges results across sheets when a table spans multiple sheets.
Sheet Selection with Start/End Page
The Start Page and End Page settings in the settings modal map to sheet numbers when processing Excel files:
- Start Page = 1 means start from the first sheet.
- End Page = 3 means stop after the third sheet.
- End Page = 0 (default) means process all sheets.
This is useful for workbooks with many sheets where only a few contain the data you need.
Differences from PDF Extraction
| Aspect | Excel | |
|---|---|---|
| Page unit | Physical page | Worksheet/sheet |
| Text extraction | X2Text adapter renders the page | X2Text adapter reads cell data |
| Table boundaries | Detected visually by the VLM | Table structure is inherent in the spreadsheet format |
| Cross-page merging | Merges rows split at page breaks | Merges rows across sheets if the same table continues |
| Formatting | Depends on PDF rendering quality | Cell values are read directly, preserving data types |
Tips for Excel Documents
- Name your sheets descriptively. Sheet names appear in the agent's context, helping it identify relevant sheets.
- Use Start/End Page to target specific sheets if your workbook has many sheets but only a few contain the target data.
- Avoid heavily merged cells. While the extractor handles merged cells, extensive merging can reduce accuracy.
- Keep formatting clean. Remove unnecessary blank rows/columns around the table — they can confuse the table boundary detection.
- Multi-sheet tables work best when the table structure (column headers) is consistent across sheets.