Train your knowledge

Connecting your first sources to Living Knowledge

Connecting your first sources is the moment your Unless agent starts to become useful. Before that, it has nothing to ground its answers on, no matter how well you configure the workspace or install the snippet. The quality of your answers depends entirely on the quality and structure of the sources you feed in, so choosing the right starting point matters more than adding many sources at once.

What counts as a source

A source points at the documents, products, and pages the agent should learn from. The Knowledge base ingests content from several categories, and you can mix them freely:

• Website pages - Crawl a full site through sitemap.xml or add individual URLs. Include and exclude rules narrow the scope.
• FAQ entries - Question-and-answer pairs you write directly or import as CSV. Best for content that does not already live on a page.
• Files - PDF, DOCX, Markdown, CSV, and Google Drive files. Structure matters more than length: clear headings, short paragraphs, no dense tables.
• Helpdesk content - Tickets and articles pulled directly from Zendesk, Freshdesk, Jira Service Management, Zoho Desk, and similar.
• Knowledge bases - Confluence, Google Drive, BookStack, Zenya, ClickHelp, Azure DevOps Wiki, WordPress, and similar.
• Team collaboration - Support tickets and developer merge requests can feed Living Knowledge when the most accurate institutional knowledge lives there.
• MCP servers and custom connectors - For internal systems on a custom protocol.

Choosing your first source

Pick the source that answers the most customer questions today. This is usually your help center, your main marketing site, or your product documentation. A single well-structured source beats several noisy ones, because the agent retrieves relevant chunks at conversation time and uses them to ground its answer.

User-generated content such as community forums or old support tickets is a poor source. It carries noise and outdated context, so skip it when you are starting out.

Adding a source step by step

1. Go to Train > Knowledge base > Sources in your Unless dashboard.
2. Add the URL of your chosen source.
3. Unless will crawl it, process it into the Content library, and make it available to the agent.

For integrations like Google Cloud Storage, Help Scout Docs, or Stack Exchange, go to the Integrations tab of your dashboard, select Configure for the integration you want, and connect your account in the popup that opens.

Understanding source status

Each source displays its connection status in the Sources list:

• Connected - The source is actively connected and synced
• Syncing - The source is in the process of updating
• Error - There was a problem connecting to or reading from the source
• Disconnected - The source is not currently connected
• Team only - The source is restricted to team members only

Improving source quality

Each source carries a Quality score that shows how well it reads for the agent. Sort by quality from low to high and start with the worst. You can also use the Content optimizer add-on, which improves source formatting such as chunking and headings without changing the original content. Your account manager can enable it on request.

What happens after you connect

Once a source is connected, the Engine starts building Living Knowledge. The content library captures the structured knowledge that emerges from ingestion, showing content nodes generated from your sources. From there you can move on to the Content library to see how your sources are structured, set up Procedures for actions, segment visitors with Audiences, and preview the agent before customers do.

Keeping sources current

Living Knowledge maintains itself. It re-reads connected sources automatically when they change, picks up new tickets and conversations as they happen, rewrites content when the underlying truth changes, and captures every accepted override and edit. This means your first source will not go stale, as long as you keep the original content updated.

The same library feeds three surfaces: the Customer Agent, the Team Assistant, and the Help Center. One source of truth everywhere, with no duplicate content operations. Start with one good source, verify the agent answers well, then expand.