Skip to content

Supported file types

Curator reads text and metadata from common text, PDF, and image files. That extracted content is what makes a file searchable and lets chat cite it.

Any file can still be browsed by name in All Files even when Curator cannot read its content. It just will not show up in content or meaning-based search until its text can be extracted.

These are the files Curator reads all the way through, so their contents become searchable.

CategoryExtensionsHow it is read
Text and code.txt, .md, .csv, .json, .xml, .html, .yaml, .toml, .ini, .log, plus common code files (.py, .js, .ts, .rs, .go, .java, .c, .cpp, .css, .sh, .sql, and more)Read directly as text.
PDFs.pdfText PDFs are read directly. Scanned or image-only PDFs are read page by page with OCR, up to 50 pages.
Images.jpg, .jpeg, .png, .gif, .bmp, .webp, .tiff, .heic, .heif, .svg, .icoA vision model looks at the image and describes what it contains.

Curator indexes these files so you can find them by name, but it does not read inside them. It records a best guess from the filename instead, so they will not turn up in content search.

CategoryExtensionsNotes
Office documents.doc, .docx, .ppt, .pptx, .rtf, .odt, .pages, .keynoteNot read for content yet.
Spreadsheets.xls, .xlsx, .ods, .numbersNot read for content yet.
Archives and apps.zip, .tar, .gz, .7z, .rar, .dmg, .pkg, .exe, and similarSkipped entirely.
Audio and videoCommon audio and video filesSkipped entirely.

For how this content is turned into search, see How processing works.