Add documents and import website pages

Copy for agent Copied

Use document import when an article or analysis needs exact facts, examples, service details, or source pages that are not already in PromptEye. Knowledge keeps the text it can read from a source. It does not preserve the original visual layout as a document editor would.

Choose a destination first

From Knowledge → Catalogs, open a folder or stay at the top of a base, then choose Add documents. The dialog shows the destination and lets you choose another folder or base. New material can go into the project base or, if available and you have write access, the Shared base.

You can also add documents while choosing sources in the Content article wizard. That quick entry opens a document dialog from the project base. To put a source in the Shared base, organize it from the main Knowledge page and move it there.

Before uploading a batch, you can assign one content type to all files. The type describes the kind of evidence inside the documents, such as hard numbers, case studies, scope and audience, differentiators, or source pages. You can change a document’s type later. Leave it unset if you are not sure; a guessed type can make the folder’s checklist misleading.

Upload a file

The current importer accepts Markdown, text, CSV, Word DOCX, Excel XLS/XLSX, and PDF files, up to 20 MB per file. Select multiple files or drop them in the upload area. Files are read and added one at a time, with a queue showing progress and read quality.

PromptEye extracts text in your browser and stores the extracted text for Knowledge. Word layout, spreadsheet formatting, page design, images, and graphics are not kept as Knowledge text. Spreadsheet extraction includes sheet names and cell values. A PDF is read from its text layer; a scanned page that consists only of an image cannot supply text this way.

The queue reports whether the import was read fully, partially, or not at all. Open the preview/read report when it flags a problem. If useful information is embedded in skipped charts or images, add those facts in Edit text. If a PDF is scanned and has no text layer, use an OCR/text version or paste a reviewed transcription. PromptEye does not create text from the scan.

The uploaded original file is not the Knowledge source used by article generation. The extracted text is. Downloading/exporting a Knowledge document therefore gives you its readable text representation, not the original Word, spreadsheet, or PDF layout.

Resolve a duplicate name

If the destination already contains a document with the same name, choose what should happen before the new document is saved:

Replace updates the existing Knowledge document with the newly read text and raises its version number. Briefs that already use the document keep pointing to it and use the new text. A manual correction to the old text is cleared because it described the previous file.
Keep both gives the incoming document a distinct name.
Skip leaves the existing document unchanged and discards the incoming copy.

Review the destination and version in the queue before closing it. If an upload is interrupted, its queue belongs to the current browser session; unfinished files can be shown as stopped when you return. The queue is not an independent background backup of the files on your device.

Paste text

Choose Paste text, give the document a useful name, and paste plain text. Formatting is not retained. The character count appears before you add it, and the saved text can then be selected as a Knowledge source. Use this option for source material that cannot be imported cleanly from a file, or for corrected text from a scan.

Import from a website

Choose From a website and enter either a domain or a page URL. The import has two stages so you can review the addresses before PromptEye fetches text:

Choose This page to import only the URL you entered, or This page + subpages to discover pages under that path. For subpages, the importer checks the site’s /sitemap.xml and /sitemap_index.xml; it does not crawl ordinary page links to discover the whole site.
Set a page limit (10, 25, 50, or 100) and optional path or URL exclusions. The limit caps the discovered candidate list before you select pages; it does not guarantee that every selected page will be fetched successfully. Exclusions are case-insensitive and support * wildcards, for example /blog/* or */terms/*. Select Find pages to see the candidate list.
Review and uncheck any pages you do not want. Choose Import selected to fetch their page text and add it as documents. The page reader extracts page text and aims to omit navigation and footer material; review the result because extracted content can still be incomplete or include text you do not need.

The importer is for website page text. It is not a way to download and read linked PDF files. If a sitemap lists a PDF, upload that PDF directly through Upload files so the PDF text extractor can read its text layer. A scan still needs OCR or a manual text correction. Discovery only looks for the sitemap at those two standard addresses; it does not follow links between pages or use a sitemap added elsewhere in PromptEye.

Why are no subpages found?

Subpage discovery depends on a usable sitemap XML file at the site’s standard sitemap locations. If the site has no sitemap, the URL is wrong, or the sitemap does not list pages beneath the path you entered, the list can be empty. The empty list does not say which of these is the cause. Try This page for one known URL, check /sitemap.xml and /sitemap_index.xml, or ask the website owner to publish a sitemap XML file. A sitemap helps list candidate URLs; it does not guarantee that every listed page can be fetched or has readable text. Pages that cannot be read are skipped without a separate error list, so the final imported count can be lower than your selection. Import the missing pages again with This page.

Why did an imported page have little or no text?

Check that the address is a public HTML page and that it contains readable page text. The importer may fail if the page blocks access, needs a login, redirects somewhere unexpected, or has content rendered in a way the reader cannot retrieve. Use the individual page URL, narrow the import to This page, and inspect the document preview after import. If the site publishes the relevant information only in a PDF, upload the PDF itself instead of importing its link as a web page.

Review and correct the extracted text

Open a document preview to check the exact text that will feed later work. The read report distinguishes text successfully extracted, partial reads where graphics were skipped, and files with no readable text. It can also show manual edits and replacement versions.

Choose Edit text to correct extraction problems or add facts that were present only in images. Saving updates the displayed text and edited character count. Note that the article writer uses the text read from the original upload, not your edits. If the corrected text must inform an article, paste it as a new Knowledge document or upload a corrected source file, then select that source in the article wizard. The original character count remains in the read report as a record of what the file reader extracted.

Set a content type after reviewing the text, and put the document in the folder that best matches its subject. These steps matter because a linked folder’s material checklist uses readable text and content types, while an article writer needs the right sources for its prompt.