Skip to main content
Libra takes far more than PDFs. Office documents, images, emails, spreadsheets, plain text and audio recordings all upload into Project files, and the text in each one is extracted so it can be searched, answered from, and cited. What differs between formats is the preview. Some files open in the full PDF viewer, some render as an inline image or a plain-text pane, and a few upload and index perfectly well without ever opening in the viewer.
Libra mascot excited
Word files convert to PDF for preview, but the original is preserved; Download always returns what you uploaded, never the converted PDF. Handy for handing back source.

What Libra accepts

Every format below uploads, is indexed and opens in the document viewer. The last column says what its preview looks like. A further set of formats uploads and indexes but has no preview of its own: .tiff / .tif scans, and a long list of plain-text and source-code extensions (.rst, .ini, .conf, .sql, .py, .js and similar). Their text is searchable and citable; they just don’t open in the viewer.
.svg and .bmp are not accepted. Drop either into an upload and it is skipped as an unsupported file type. Convert to PNG first.
Files are capped at 500 MB each and 1,000 per chat. Very large spreadsheets are the one case where a file well inside the size cap can still be refused during processing; the FAQ has those bounds.

Format-specific notes

Each family has its own habits, in the viewer and in how Libra reads it.
The richest preview. Page navigation, zoom and citation jumps — arrive from an answer and the cited passage is highlighted for you — plus the Anonymize action in the toolbar, and Translate where your subscription includes it.
Word files are converted to PDF on upload, so the preview behaves identically to a PDF — including page controls, zoom and anonymisation. The original is preserved; Download returns the editable file you uploaded.To edit a Word document in place with AI assistance, use the Word Add-in.
.pptx, .ppt, .ppsx, .pps and .odp decks convert to PDF, as do .odt documents and .ods spreadsheets. Once converted they behave like any other PDF in the viewer, and the original stays available for download.
An .xlsx workbook opens as a read-only spreadsheet: the grid with its sheet tabs, formulas worked out, and the workbook’s own charts drawn as Excel draws them. You can scroll, switch sheets and select cells, but not edit.A very large workbook, over 500,000 cells, falls back to the earlier preview. So does an older .xls: it is rendered as one scrolling page, with each sheet under its own heading.Behind the preview, Libra reads each sheet as a table and finds the header row even under a title line. A citation shows the cited rows as a small table labelled with the sheet and cell range, such as Offene Posten · A4:F7.A .csv opens in the same spreadsheet view, as one sheet named after the file, and the File type filter in Project files groups it under Excel.Values stay as they are in the file: a cell that starts with = is shown as text, not run as a formula. For search, answers and citations it is read as a table, like a workbook.Comma, semicolon, tab and pipe separators are all recognised, so a semicolon export from a German accounting system indexes like any other..csv and .xls files uploaded before 23 September 2026 keep their earlier, flat-text indexing until you upload them again.
Rendered inline and scaled to fit the pane. There are no zoom or pan controls — zoom is a PDF-viewer feature.Text in images is extracted automatically, so a photographed or scanned page is searchable and citable once processing finishes. .tiff scans are text-extracted the same way but have no inline preview.
The viewer shows the Subject, then From, To, Cc and Date, then the message body. Where the email carries an HTML part, that part is rendered — sanitised, with inline images embedded and any external references stripped. Plain-text-only emails render as text.The attachments of an .eml are read as if you had uploaded each one, so a scanned PDF is text-extracted and a workbook is read in full. Libra searches and cites them with the message, though the preview shows only the message.An Outlook .msg file opens the same way, with its body shown as plain text that keeps its paragraphs, lists and table rows. Its attachments are left out, so upload the ones you need as their own files.
.txt, .md, .log, .xml and .json render as monospaced plain text, with no syntax colouring and no formatting. Useful for transcripts, exports and logs.For anything you want to analyse rather than read, point a Discovery at the file instead of opening it here.
.html and .htm files are sanitised on the server — scripts and styles stripped, external resource loads blocked — and rendered in a sandboxed frame. Useful for archived web pages and saved articles.When Libra reads the page, its headings, lists and tables keep their structure, so an answer can draw on a whole table row.
Audio uploads are transcribed, and the transcript is what Libra searches and cites. There is no player in the document viewer.A recording can run to four hours. A longer one fails to process with Audio exceeds the 240 minute limit; split it and upload the parts.
Email, .xls and HTML previews are generated on the server at the moment you open them, and that generation is capped at 10 MB.A larger .eml, .xls or .html file uploads and indexes normally, but the preview itself fails to load. Search, answers and citations still work on it.

Working in the viewer

The tab strip and the Download button are there whatever the file is. The page, zoom, Translate and Anonymize controls appear only on the PDF path. Tabs. Every document you open becomes a tab along the top. Click a tab to switch, drag a tab to reorder, and click the × on a tab to close it. Once more than one tab is open, arrows appear for stepping through them. Opening an answer’s sources adds a single Citation list tab at the front of the row — it is re-pointed at whichever answer you came from rather than stacking up one tab per answer. The document viewer with two tabs open, a PDF lease summary and a PNG floor plan, the image tab active and the Project files panel still visible on the left Download. The toolbar has a Download button on every preview, and it returns the original file. A Word document downloads as .docx, not as the converted PDF. Translate. PDFs and .docx files can be translated into another language from the toolbar, up to 30 MB per file and subject to a daily limit. Translations open as their own document. Anonymize. PDFs with selectable text — and therefore anything converted to PDF — can be anonymised in one click from the toolbar. Libra opens a copy with personal data replaced by placeholders and leaves the original untouched. See Anonymise a document.

Citations across formats

An answer in Chat, a Review result or a Discovery cell can cite any indexed file, and clicking the citation opens that file in the viewer. How precisely it lands depends on the preview:

Tips

If you have the original .docx, upload that rather than a PDF you exported from it. Libra converts it for preview but keeps the original, so the download stays editable.
Scanned PDFs, photographs and .tiff files are text-extracted in the background. No hits on a fresh upload usually means extraction is still running, not that the text is missing. Wait, then re-run the query.
Libra reads an .eml’s attachments, but the preview shows only the message, and a .msg file’s attachments aren’t read at all. If an attachment is the document that matters, upload it as its own file so you can open it in the viewer.
A folder of spreadsheets is a Discovery’s natural input. Rather than paging through each workbook in the viewer, extract the fields you need into a Discovery table.
.svg and .bmp are refused outright, and .tiff gives you indexing without a preview. If you need the file on screen as well as in the index, convert it to PNG or PDF before uploading.
Libra mascot frustrated
Uploaded a mixed batch and came up short? An unsupported type doesn’t fail the whole upload — the rest goes through and a notice names what was skipped. Worth reading before you assume everything landed.

Next steps

Anonymise a document

Share a copy without the personal data.

Folder structure

Organise Project files into folders.

Project files

Back to the Project files overview.