Add content to a collection

Open Knowledge, click a collection, then Add document. That dialog offers four ways to get content in, and they all end in the same place: a draft you then publish. There is a fifth door, on its own tab, at the bottom of this page.

Text#

Paste or type the content. Best for short, hand-written answers you want full control over: opening hours, a returns policy, the three things people always ask.

File#

Upload a PDF, a Word document, an HTML file, Markdown or plain text. The text is extracted in the background, so the document appears immediately and its passages arrive a moment later. Watch the Processing tab if it is a large file.

URL#

Give a link and we fetch the page and keep its text. Use the scope field to say which part of the site the fetch is allowed to touch; anything outside it is refused. You can also ask for the page to be fetched again on a schedule, so a page you keep updating stays in step without you doing anything: switch on Check this page for changes and pick every so many hours, days, weeks or months, from one hour up to a year. Only a published document is checked, a page that has not changed costs nothing, and the document's own page shows when it was last checked and when it is due again. An interval with no next check beside it means the document is not published yet.

Your website#

Give us the site, just example.com, and we find how to crawl it: we read your robots.txt for a sitemap declaration, try the usual sitemap addresses, and if the site publishes none we follow its links instead. You do not need to know where your sitemap lives, and tick I know my sitemap address only if you do.

The whole crawl becomes one document made of many pages, so an answer still links to the exact page it came from. This is the fastest way to bring in a whole help centre. Large sites take a while, and the Processing tab shows how far along the crawl is, how many pages it found, and anything it skipped.

Leave "Most pages to read" empty and we read every page. There is no hidden limit: set a number only if you want one. If a crawl does stop short, the job says so and why.

Your own storage#

If the files already sit in object storage you own, do not upload them one at a time. Register the bucket once under Your storage, then import a whole folder as often as you like: a repeat import fetches only what actually changed. It has its own article.

Then publish#

Everything above lands as a draft. Nothing is searchable, and no agent can use it, until you publish. Open the document and click Publish.

Publishing is also what you do after an edit. If you change the text of a published document, publish it again or the agent keeps answering from the previous version. The document view tells you when the two have drifted apart.

More than one language#

A document can carry versions in several languages. Add the translation as a language version rather than as a second document, and one publish indexes them all. The agent then finds the version that matches the language of the call, which is what you want when the same policy has to be quoted in Greek and in English.

Tags and categories#

Optional labels. They are useful once a collection grows: you can filter by them when you search, and they travel with the document.