Skip to main content
Upload files to a collection in a single request. The file is stored, validated, and immediately queued for parsing and knowledge extraction.

Upload a file

Response:
The file enters the processing pipeline automatically. Check the collection status to know when extraction is complete.

Supported formats

Validation

Files are validated in three layers before upload:
  1. Extension — must be one of the supported formats above
  2. MIME type — must match the expected type for the extension
  3. Magic bytes — file header is checked to confirm actual format (prevents renaming executables to .pdf)
Executable files (.exe, .sh, .bat, .dll, etc.) and binary files (ELF, Mach-O) are always rejected.

Limits

Error codes

Processing pipeline

After upload, the file goes through the processing pipeline:
  1. Parsing — document is parsed into zones (sections, tables, etc.)
  2. Extraction — knowledge cards are extracted from zones using LLM
  3. Indexing — cards are embedded and indexed in the vector database
Reference files (.xlsx, .csv) follow a separate pipeline optimized for tabular data. The collection status changes to processing during this time and processed when complete.