Format support | Perception Docs

An operation refuses a file type it doesn't read with 422, code unsupported_file_type, when the job is created. How the document service tells the type depends on the source:

Uploaded files (workspace_file): when it's uploaded, from its extension, checked against its first bytes. A file whose bytes clearly say another type (a PDF named .bmp) is refused at upload with 422 unsupported_file_type. A file with no supported extension gets the type its bytes show, and one whose bytes are inconclusive, such as plain text, keeps its extension's type.
Links (url): from the extension of the source's file_name, since nothing has been downloaded yet. Name the file with an extension from the table below. The SDKs and the CLI take file_name from the URL when it ends in one; see Sources.

File type	Extensions	Parse	Split	Classify	Extract	Ground
PDF	`.pdf`	✅	✅	✅	✅	✅
Spreadsheet	`.xlsx` `.xlsm` `.xls` `.csv`	✅	✅	✅	✅	✅
Parquet	`.parquet`	✅	❌	✅	✅	✅
Word document	`.docx` `.doc` `.docm`	✅	✅	✅	✅	✅
Image	`.png` `.jpg` `.jpeg` `.tiff` `.tif` `.webp`	✅	✅	✅	✅	✅
Bitmap	`.bmp`	❌	✅	✅	✅	✅
Presentation	`.pptx` `.ppt`	✅	✅	✅	✅	✅
HTML	`.html` `.htm`	✅	✅	✅	✅	✅
Plain text	`.txt` `.log` `.xml`	✅	✅	✅	✅	✅
Diagram	`.vsdx` `.vsd`	✅	✅	✅	✅	✅
Markdown	`.md` `.mdx`	✅	✅	✅	✅	✅
JSON	`.json`	✅	✅	✅	✅	✅
JSON Lines	`.jsonl`	❌	❌	❌	❌	✅
Email	`.eml` `.msg` `.oft`	✅	✅	✅	✅	✅
Video	`.mp4` `.mov` `.avi` `.mkv`	✅	❌	❌	❌	✅
Audio	`.mp3` `.wav` `.m4a` `.flac` `.mpga` `.oga` `.ogg`	✅	❌	❌	❌	✅

Audio and video: Parse transcribes them. See Parse.
Markdown is read as it is, not OCR'd. Parse returns its text unchanged except for tables, which follow output.table_format: HTML by default, so pass "markdown" to keep pipe tables.
Large plain text: Parse reads .txt files up to 1 GiB and returns the complete text through document.content_url. Split, Extract and Ground read .txt files up to 256 MiB.