Skip to main content
Files that are well structured convert cleanly to Markdown and produce better training data and vector stores. Check your file format and structure before you upload files.
This page covers the app. To prepare files in code, see Prepare and ingest files in the SDK guide.

Choose a file format

The formats you can upload depend on the file purpose you select at upload. Raw Data AI-Ready Data jobs and vector stores accept PDF, Word, PowerPoint, Markdown, and text files. They don’t accept .json files. Fine-tuning Data (Instruction Tuning Data and Reinforcement Tuning Data) You can upload up to 20 files at a time, each up to 150 MB. The upload dialog flags a file in an unsupported format with Unsupported file format, and a file over the size limit with File exceeds 150 MB. Remove flagged files from the list before you upload.

Structure your documents

Structure affects how well SeekrFlow converts a file.
  • PDF and Word – Use clear headings and arrange text in a logical order. Add context around images.
  • Markdown – Use a correct header hierarchy (# H1, ## H2, ### H3). Give every section meaningful content, and avoid empty or skipped header levels.
If you write in Google Docs, export the document as Markdown, PDF, or Word.

Structure JSON files

A JSON file is a tree. Each node is a section of content and can contain nested child sections. Each node requires these fields:
  • label – A short title or question for the section.
  • content – The main body text for the section.
  • children – An array of child objects with the same structure. Use an empty array ([]) if the section has no children.
  • level – The depth of the section in the hierarchy. Use "0" for the root, "1" for a first-level child, and so on.
Last modified on November 19, 2026