# Knowledge limits and troubleshooting

Check the source type, exact size boundary, and network safety rules before retrying a failed knowledge item.

## Exact ingestion limits

- Created text: 2 MB maximum.
- Imported public URL content: 2 MB maximum.
- Uploaded file: 20 MB maximum; extracted text must also fit the text-content limit.
- Accepted extensions: `.txt`, `.md`, `.pdf`, `.docx`, `.html`, and `.epub`.
- Move or recursive bulk-delete selection: 1 to 100 entities.

A scanned PDF with no text layer is rejected because OCR is not provided. Export it as searchable text or upload an extracted text document. A corrupted or password-protected PDF also needs a readable replacement.

## URL safety failures

URL import accepts only `http` and `https` public destinations. A private address, loopback address, link-local address, or non-public DNS result is rejected. Redirects are limited and every redirected destination is checked again, so a public starting URL cannot redirect into a private network.

URL import also rejects unsupported response types and content larger than 2 MB. It does not run page JavaScript. If a browser-rendered page imports as empty, provide a direct readable page or create a text document instead.

## Status diagnosis

- `indexing`: processing has started; wait before attaching or publishing.
- `ready`: indexing completed and the document can be referenced.
- `failed`: processing did not complete. Open the item, inspect the source, and replace or retry with corrected content.

If publishing reports a missing or not-ready document, return to Knowledge Base and check each referenced item's current status. If a text or file size looks close to the limit, reduce it below the exact boundary rather than retrying the same payload.

For ingestion steps, return to [add sources](/docs/knowledge/add-sources). For deletion effects, read [manage and delete](/docs/knowledge/manage-and-delete).