Keep the sheet name
The sheet name is part of the meaning. 'Q3' and 'Assumptions' are not the same table even when a column header is reused.
Record the sheet as a page or an equivalent block group so a citation can name it.
Headers stay with the rows
The first row is often the schema. Each following row is a record. Chunks should repeat the header, the way a long PDF table repeats its header.
Do not embed the entire workbook in one model window. Retrieve the sheet and the rows that match the question.
Formulas and displayed values
The XLSX path loads the workbook with data_only set to false, so a formula cell is stored as the formula text, not the last calculated number shown in Excel.
A sheet with no non-empty cells adds the warning Sheet had no non-empty cells, followed by the sheet title. Do not index that sheet as a successful blank document.
What one workbook becomes
Each worksheet is a DocIR page. A hidden sheet is skipped and the warning says Hidden sheet skipped, followed by the sheet title. A visible sheet starts with a level-1 heading Sheet: and the sheet title, then one table block of the non-empty rows.
The header is the first row of that table, not a separate schema object. Chunks of a long sheet should repeat that row. The default route for a text-based XLSX is container.
A hidden sheet still receives a DocIR page, and that page has no table. The warning is Hidden sheet skipped, followed by the sheet title. The next visible sheet keeps the later pageIndex, so pageIndex is not a count of visible sheets. An empty visible sheet adds the warning Sheet had no non-empty cells, followed by the title, and that page also has no table. A formula cell is the formula string because the workbook is loaded with data_only set to false. Legacy XLS and XLSB need LibreOffice first. If LibreOffice is unavailable, that conversion fails closed.
- Cite the file and the Sheet heading, not a PDF page number.
- Read formulas as formula text.
- Skip a sheet whose warning says it was hidden.
- Repeat the header row on every chunk of a long sheet.
The file to send
Send one XLSX. The default route is container. Each worksheet gets a page before the hidden-sheet check. A visible sheet starts with a level-1 heading Sheet: and the sheet title, then one table. The first row is the header.
Formula cells are the formula text. A hidden sheet adds the warning Hidden sheet skipped, plus the title, and still consumes a page index. XLS and XLSB need LibreOffice first and fail closed when it is missing.
- Read the Sheet heading in result.markdown.
- Cite the sheet title and the row, not a PDF page number.
- An empty visible sheet warns Sheet had no non-empty cells.
export DOCPARSE_URL="https://docparse.genedai.me"
export DOCPARSE_API_KEY="your_key"
curl -fsS -X POST \
"$DOCPARSE_URL/v1/parse?wait=true" \
-H "Authorization: Bearer $DOCPARSE_API_KEY" \
-F "file=@workbook.xlsx" \
-F "mode=both" \
-o result.json
jq -r '.result.markdown' result.json
jq '.result.document.schemaVersion, .result.document.pages[0].blockIds' result.json
jq '.result.manifest.decision.route, .result.manifest.parserVersion' result.json
Frequently asked questions
Questions this guide answers
Does DocParse accept XLSX?
Yes. XLSX, XLSM, XLTX, and XLTM are routed through POST /v1/parse. The outputs are Markdown and docir.v1. Legacy XLS and XLSB need a LibreOffice conversion first.
Should each sheet be its own document?
Keep the workbook as one source. Each visible sheet is a page whose heading is Sheet: plus the sheet title, so a citation can name the file and the sheet.
How do I chunk a large sheet?
Repeat the first row, which is the header, on each row group. Keep the sheet heading and the source hash on the chunk. Formula cells are the formula text, because the workbook is loaded with data_only false.