Supported File Types

---

Document formats

Format	Extensions	Notes
PDF	.pdf	Native rendering in viewer. Text, tables, and images extracted. Scanned pages are OCR'd automatically.
Word	.docx	Rendered in browser. .doc (old format) not supported — save as .docx first.
PowerPoint	.pptx	Converted to PDF for preview via LibreOffice on the server. .ppt is converted and parsed only if LibreOffice is installed on the server; otherwise it indexes with no extracted content.
Excel	.xlsx, .xls	Tabular data extracted and queryable via natural language.
CSV	.csv	Same as Excel — rows are queryable via natural language.
Plain text	.txt	Indexed as-is.
OpenDocument Spreadsheet	.ods	Same as Excel.
OpenDocument Text	.odt	Not supported — save as .docx first.
Images	.png, .jpg, .jpeg, .heic, .heif	OCR extracts printed text — scanned pages, photographed documents, signage. See "Scanned documents and handwriting" below.

---

File size limit

500 MB per file. For files near the limit, indexing may take 3–5 minutes.

---

Tips for best results

PDFs:

• Text-based PDFs (not scanned images) give the best accuracy
• Scanned PDFs work but require OCR — accuracy depends on scan quality. Correct page orientation before uploading; sideways or upside-down scans are not reliably extracted.
• PDFs with complex tables are handled well; heavily formatted multi-column layouts may lose some structure

Images:

• Sharp, flat, well-lit photos of printed pages read well; screenshots of documents also work
• Low-resolution, rotated, skewed, or glare-affected photos may extract poorly or not at all
• If an image contains no readable printed text, the document indexes with a warning ("No extractable text found") and will not contribute to answers

---

Scanned documents and handwriting

BodhAI OCRs printed text on scanned (image-only) pages and on uploaded image files, using a local OCR engine — image content never leaves your server. Extraction quality depends on scan quality: clean and photocopied printed scans read well, while rotated, skewed, or very low-quality pages may not extract reliably.

Handwriting is not supported. This is a permanent product boundary, not a "coming soon" feature — BodhAI does not perform handwriting recognition, and there is no plan to add it. Treat any handwritten content on a scanned page as not extracted. If something is returned for a handwritten region, do not trust it.

Handwritten numbers are the sharpest risk — verify them by hand. A misread handwritten digit produces a valid-looking but wrong value with nothing to flag it as suspect: a "7" read as a "1", or a "3" as an "8", becomes a plausible-looking quantity, price, measurement, batch code, or date that can flow silently into an answer. Never rely on BodhAI to read handwritten numerics — always verify handwritten figures against the source document yourself.

Excel:

• Column headers are important — clear, descriptive headers improve how BodhAI interprets queries
• One sheet per workbook gives better results than many sheets in a single file
• Remove purely decorative rows (merged title cells, blank spacers) before uploading

PPTX:

• Slide titles and bullet points are indexed. Dense text slides work better than purely image-based slides.
• Speaker notes are also indexed if present

---

Formats not supported

• .doc, .odt (old/other Word formats) — convert to .docx first
• .ppt — parsed only when LibreOffice is installed on the server; otherwise indexes with no extracted content
• .tiff images — convert to PNG or JPEG first
• Handwriting — not recognised (printed text only; see "Scanned documents and handwriting" above)
• Audio and video files
• Zip archives — extract and upload individual files