The Knowledge Sets Manager lists the reusable knowledge bases your agents retrieve from, and is where you create, configure, publish, and share them.

Overview

A knowledge set is a reusable knowledge base — the "set" that gives agents accurate, grounded answers. Inside a knowledge set you add one or more knowledge sources (documents, websites, connected apps, or data catalogs), and the platform indexes them so agents can retrieve relevant pieces at answer time. Because a knowledge set is its own asset, you can build it once and attach it to several agents. The Knowledge area lists every set in the workspace.

Creating a Knowledge Set

On the Knowledge list, use New Knowledge Set (the empty state offers the same as Create Knowledge Set). The dialog — "Help AI make more informed decisions by adding knowledge sources" — asks for:

On create you land in the set's Configuration tab to add your first source. A new set starts from your workspace's global indexing settings (embedding model, chunking), which you can override later.

What's on a Knowledge Set's Details Page

A knowledge set opens on two tabs: Configuration and Analytics.

• Configuration is where you build the set — a side-nav lists your sources grouped by type (Application, Documents, Websites, Data catalog) and the set's settings: Indexing (how content is parsed, chunked, and stored), Enrichment, Redact sensitive information, and Storage.

• Analytics reports on the set's content and indexing over time.

You add sources from the New Knowledge button — see Add Knowledge Source.

What's on the Knowledge Set List

Each row shows the Knowledge Set Name, Tags, Owner, and Last Updated; a Workspace column appears when viewing across projects. Search with Search knowledge sets. Each row's actions are:

If you have none yet, the empty state reads Create Knowledge Set — "Help AI make more informed decisions by adding knowledge sources."

How a Knowledge Set Is Published

A knowledge set is a versioned asset, like an agent. As you add sources and change settings you're working on a draft — the set shows Undeployed Changes until you Publish a version. Publishing makes the current configuration the deployed one; the set keeps a version history you can open from its overflow menu.

Note: Publish after you've added and indexed your sources so agents use a stable version.

How Agents Use a Knowledge Set

Agents retrieve from a knowledge set at answer time — you attach the set to an agent from the agent's Knowledge section. There's no separate "test retrieval" screen on the knowledge set itself; you validate retrieval by asking the attached agent questions and checking its answers (and its trace, which shows the retrieved chunks). Retrieval tuning such as re-ranking lives on the agent, not the knowledge set. At question time the platform scopes the search to the agent's attached sets and sources and filters by the chunk's set/source identifiers stored in the index — so one shared index serves many agents without cross-talk.

Disabling, Deleting, and Re-syncing Sources

• Disable (the row toggle) — the source stops being retrieved: disabled sources are excluded from the agent's searches. The source returns to Draft; re-enabling triggers a sync.

• Delete — the source's chunks and embeddings are removed from the knowledge store, along with its staged data and sync state. This is permanent.

• Re-sync — each sync writes a new version of the source's chunks; retrieval uses the latest version, and superseded chunks are cleaned up after the sync completes, so answers never mix old and new copies of the same document. Content that disappeared at the origin is removed from the index on sync.