Connect your knowledge

Before you start

• Difficulty: Easy
• Time: 10 minutes
• You need: An admin or owner account
• You need: For a connector: its access details, such as a Notion integration secret, a Zendesk or Confluence API token, or an Intercom access token

Steps

1. Add a website

   In Knowledge sources, add the site address. ZChat reads the sitemaps it finds (robots.txt Sitemap lines, sitemap.xml and sitemap indexes), then follows links on the same host. Include and exclude patterns narrow what is read: * matches anything, a pattern without * matches as a prefix, and exclude wins over include. The first sync starts when the website is added.

2. Connect a tool

   Choose Notion, Zendesk Help Center, Confluence Cloud, Intercom Articles or Web pages & RSS, enter its details and save. The stored secret is encrypted.

3. Upload a file

   On Knowledge Base, upload a file of up to 10 MB. Accepted types are .txt, .md, .markdown, .csv, .docx, .pdf, .pptx, .xlsx, .html, .htm, and the images .png, .jpg, .jpeg and .webp. Each file is imported as an article with the status Draft. The article is switched on when it is created, so the AI can answer from it, and a draft article has no public link.

4. Pin an exact answer

   On Knowledge sources, under Pin an exact answer (Q&A pair), enter a question and its answer and add it. The pair is saved as a knowledge article keyed by its question.

What each source imports

• Website: The site address, optional include and exclude patterns, and up to 10 website sources per workspace. Up to 60 pages per sync for each website.
• Notion: An integration secret. Imports the pages and databases shared with that integration. A page published to the web keeps its public address; a private page is cited by its title.
• Zendesk Help Center: Your subdomain. Public articles need nothing else. An agent email and API token also import restricted articles, which are cited by title only. Draft articles are skipped.
• Confluence Cloud: Your site, an account email and an API token. Pages in spaces open to anonymous visitors are cited with their link; other pages are cited by title only.
• Intercom Articles: An access token and your region (US, EU or AU). Imports published Help Center articles.
• Web pages & RSS: Up to 50 public page or feed addresses. A page becomes one article, and each feed item becomes one article.
• Files: Up to 10 MB each. Each file is imported as an article with the status Draft.
• Q&A pairs: A question and its exact answer, saved as one article.

Good to know

• Limits: 60 pages per sync for each website, 10 website sources, 100 documents per sync for each connector, and 200 articles per workspace across all sources.
• A sync that fails outright changes nothing, so an outage or a revoked token does not empty the knowledge base. The source shows the error under Last sync problem.
• Disconnecting a source removes all of its articles.

What you will see

Each source card shows counts for its articles, and for the items of its last sync: indexed, skipped, failed and removed. It also shows the last complete and partial sync times and when the next sync is due.

• Articles, indexed, skipped, failed and removed counts for each source
• A Last sync problem line when a sync failed or could not read every item; a complete sync shows no such line

How a retired page is handled

ZChat removes an article only when the source reports it gone or a complete listing no longer includes it. Each source type works as follows.

• Website: a page that answers 404 or 410 is switched off, so the AI stops quoting it. If the next sync also gets 404 or 410, the article is removed after two syncs in a row. Any other answer, such as a timeout or a server error, leaves the article as it is, and a page that answers normally again is switched back on.
• Notion: a page that Notion reports as archived or in the trash is removed in that sync.
• Zendesk, Confluence, Intercom and Web pages & RSS: in a sync that reads every item without a failure, an article that is no longer listed is removed in that same sync.
• A source with more items than one sync reads: the held articles that the sync did not list are checked one at a time. A 404 or 410 answer switches an article off, and a second one in a row removes it. Any other answer keeps the article.
• A sync that cannot read every item removes nothing. The items that were read are updated, the others are kept, and the sync is marked incomplete so that you can retry it.

Common mistakes

• Reading a source with a Last sync problem as a full sync. Some items were not read, and that run removed nothing.
• Expecting a restricted Zendesk or Confluence page to be quoted in full. The AI cites it by title only.
• Expecting a Notion page to import without sharing it with the integration. Only shared pages are listed.
• Expecting an uploaded file to stay out of answers until you publish it. The file is switched on when it is imported. To keep it out of answers, switch its article off in Knowledge Base.
• Expecting an edit made in the source to reach the AI at once. Changes reach answers at the next sync of that source.