Add files, links and recordings to a knowledge base

Purpose

The return terms are in a PDF, the delivery rules are on a web page and the product catalog is a spreadsheet. Neda wants all of it in one place without retyping, and this section shows how to add each kind of source and follow it until it is ready to be searched.

Before you start

Open a knowledge base (see "Create and manage a knowledge base"), or a project's Knowledge tab. You need permission to manage projects to add or remove items. Check the limits of your plan: they cap the number of items, the indexed text, the scanned pages and the transcription minutes.

Steps

1. In the Sources part of the page, drop one or more files onto the box Drop files here or browse, or select Choose files and pick them. You can add several files at once. The box accepts PDF, Word (.docx), text, Markdown and HTML files, spreadsheets (CSV, TSV, XLSX and JSON), images (PNG, JPEG, WebP, GIF) audio (MP3, M4A, WAV, OGG, OPUS, WEBM, FLAC, AAC) and video (MP4, MOV, MKV, AVI, M4V), whose sound is transcribed when possible. A file can be up to 25 MB. A recording can be uploaded up to 300 MB, but only recordings up to 25 MB can be transcribed, and images over 20 MB cannot be read. While files upload, the button reads Uploading 1 file; afterwards a note says how many files were added.
2. To add a web page, type its address in the Add a web page field (for example https://example.com/article) and select Add link. The address must be a public http or https address; otherwise the message Enter a public http or https address. appears. Addresses with a user name or password, and addresses on private or local networks, are added but then fail with This address cannot be read.
3. In a project's Knowledge tab, files that were already in the project before its knowledge base existed can be brought in with Import project files. A note says how many were queued ([n] project files queued). New files added to the project's Sources join the knowledge base by themselves.
4. Every file or link becomes an item in the Items list, with an icon for its kind, its name, size, page count or length in minutes, the date it was added and a status chip: Waiting (queued), Processing, Ready (searchable), Failed or Skipped. Spreadsheets are read row by row, scanned pages and images are read as pictures, and recordings are transcribed. The page updates by itself while anything is waiting or processing. Before anything is added, the list says Nothing here yet. Add files or a link to get started.
5. Above the list, a progress bar shows [n] items are still being processed and [done] of [total] items finished, with the number of failed items in brackets. When everything is finished it says Everything is indexed. The chips All, Waiting, Processing, Ready, Failed and Skipped, each with a count, filter the list; only statuses that currently have items are shown. If a filter has no items, the list says No items with this status.
6. For a failed or skipped item, a short red reason appears under its name (see the table below). To act on any item, open its options button (Options for [name]):
   • View extracted text (only for Ready items) opens Extracted text, the text the knowledge base actually indexed. Very long texts show only the first part and say so. If nothing was extracted, it says No text was extracted.
   • Open original opens the stored file; for a web page it reads Open link.
   • Process again reads the item once more, for example after a limit was lifted. It is disabled while the item is Waiting or Processing.
   • Delete asks Delete this item? and then removes the item and its indexed text.
7. When the list is long, more items load as you scroll. If they cannot load, you see Items could not be loaded. with Retry.

Expected result

Each source appears as an item that moves from Waiting to Processing to Ready. Ready items can be searched with Test search and are used when you ask about the knowledge base in a chat. Items that could not be read show a plain reason.

Troubleshooting