The Add Knowledge Source flow is where you connect one input — documents, a website, an app, or a data catalog — to a knowledge set and configure how it stays current.

Overview

A knowledge source is one input to a knowledge set — a set of documents, a website, a connected app, or a data catalog. You add sources from a knowledge set's Configuration tab: the New Knowledge button opens a menu of the source types, and picking one opens that type's editor directly. Each source gets a name, a description, and tags, and then type-specific settings for where its content comes from and how it stays in sync.

Source Types

There are four source types. You can add several sources of different types to one knowledge set.

Pick the type that matches where your content lives.

Importing a Website

Choose Import a Website and decide the scope:

• All webpages of a URL — crawl from a starting URL. You can include linked documents, and cap the crawl with a Webpage fetch limit (default 100) and a Crawl depth (default 2).

• Specific webpages — list exact pages, separated by commas.

Set a Refresh Frequency — Daily, Weekly, Monthly, or Yearly — to re-sync the site automatically so the knowledge stays current.

Behind the scenes the crawler renders each page, converts it to Markdown, and indexes it, then follows the links found on that page — skipping pages it has already visited — until it hits your fetch limit or crawl depth. Specific webpages indexes only the listed pages (no link-following). With Include documents on, files linked from pages (such as PDFs) are downloaded and indexed through the regular document pipeline. Every fetched and failed URL is tracked per sync, which is what the source's Crawled, Pending, Synced, and Failed counters report.

Adding an Application Source

The connector's import operation defines everything about the sync: what gets downloaded (files, messages, pages, tickets…), which metadata is kept on each chunk (author, dates, file name), how source permissions are fetched, and the default incremental-sync cadence. The initial sync downloads historical content in stages (you'll see progress counters grow as items are discovered and indexed); after that, incremental grabs pick up new and changed items on a schedule, and sources that support it can switch to real-time updates via webhooks.

For permission-aware sources, each item's access list is captured during sync and enforced at question time — users only retrieve chunks from items they can access in the source app, unless you enable Skip User-Level Source Permission Checks. You can only set this when you add the Application source; afterwards it is shown read-only. To change it, add the source again. When you fill in the connector's inputs, any lookup field lets you pick from the fetched options or choose Enter custom value to supply one that isn't listed — useful when the exact site, folder, or channel you need doesn't come back in the list.

Adding a Data Catalog Source

A data catalog source doesn't create text chunks — it indexes your tables' metadata (schemas, columns, relationships) plus any curated example queries and instructions. Agents then answer data questions by generating SQL against the cataloged tables and running it, rather than by matching text.

Note: Add a data catalog source when you want an agent to answer quantitative questions from live data; add document/website/app sources for narrative content.

Keeping a Source in Sync and Controlling Retention

Each source has General settings for sync and retention:

• Knowledge Retention — turn on to set a Retention Period (days, months, or years) after which content ages out.

• Schedule Knowledge Sync — set a sync schedule (an interval in minutes/hours/days/weeks/months, or a custom schedule). The source shows its Frequency, Last synced, and Next scheduled times.

• For Application sources you may also see Skip User-Level Source Permission Checks ("Allow users to access all application knowledge regardless of their permissions on the source") and, where supported, an option to update on real-time changes. Skip User-Level Source Permission Checks is read-only here — it can only be set when you add the source.

With Knowledge Retention on, content older than the retention period is removed from the index automatically. Real-time updates, where a source supports them, work by deploying a webhook listener for that connection — the app pushes changes and only affected items are re-indexed, instead of waiting for the next scheduled sync.

Indexing One Source Differently from the Rest

By default a source uses the knowledge set's (and workspace's) global indexing settings, but you can override them per source — a note explains: "Indexing settings will use the global agent configuration by default. You can customize settings here to override global defaults for the specific Mime Format. Any changes made here will only apply to this knowledge source."

Use a per-source override when one file type (say, scanned PDFs or spreadsheets) needs different parsing or chunking than everything else in the set.