Agentforce Data Library: Setup, Sources and Troubleshooting

Hey everyone,

You build an agent, ask it about your company’s return policy, and it gives a vague, generic answer. It doesn’t know your documents yet.

An Agentforce Data Library fixes that. It turns your Knowledge articles or uploaded files into searchable content an agent can answer from. You create it in Setup, pick one data source, wait until it’s ready, then add it to your agent in Agentforce Builder.

Here’s what to check before you start, how to choose a source, and what to look at when the agent still can’t find anything.

What does a data library actually do?

It handles the plumbing for grounding (giving the AI your real data to answer from). When you save a library, Salesforce automatically creates data streams, a search index and a retriever in Data 360 (Trailhead).

In plain terms: your content gets split into small pieces (chunking), stored so it can be searched (the search index), and a retriever finds the right pieces for each question.

It’s also required setup for the standard Answer Questions with Knowledge agent action, and you can use it in custom prompt templates (Salesforce Help).

Can I use one in my org?

Check three things first. If any one is missing, you won’t get far.

• Edition: Enterprise, Performance or Unlimited, with an add-on such as Agentforce for Sales or Service, Einstein for Platform, or Agentforce Foundations.
• Permissions: Data Cloud admin and System Administrator, to set libraries up.
• Data 360: it has to be set up, and data libraries consume Data 360 credits.

All three come from the Use Salesforce Knowledge Help article. If any one is missing, the setup screens stay locked.

Should I use Knowledge articles or uploaded files?

Pick carefully. After you choose a library’s source, you can’t change it later.

Help also lists web search as a source type. I’ve kept this post to the two that Trailhead walks through.

How do I set it up?

From Setup, search for Agentforce Data Library, click New Library, and give it a unique name and an optional description. Save.

For files, click Upload Files, pick files within the size limits, and save. For Knowledge, open the Knowledge tab and choose your fields (more on that next).

Then wait. Monitor progress from Setup. When the status says “Your data library is ready to use!”, you can test it with agents.

Which Knowledge fields should I pick?

This choice decides how good your answers are. There are two kinds of fields:

Identifying fields help Agentforce find the right article. Pick a field that’s a concise outline of the article, like a summary. They must be text or text area fields and are limited to 512 tokens.

Salesforce’s tip: use a simple text field. It’s capped at 255 characters, so you know you’re under the token limit.

Content fields add the details to the answer: troubleshooting steps, product details, special considerations.

Two more settings matter. Show sources adds citations, which link the answer back to its Knowledge article. And Filter by Knowledge Data Categories narrows the library to selected categories, which Help says makes indexing and grounding more accurate on a large knowledge base.

One gotcha with identifying fields: the developer guide says they can’t be changed after the library is created. Choose once, choose well (Agentforce Developer Guide).

How do I connect it to my agent?

Open your agent and click Open in Builder. On the Knowledge tab, select the library (or click New Library to make one there). Save, then Activate.

Before you test, check that the search index and the library are both ready. Then restart the preview conversation so the agent picks up the change.

New to the builder? My Agentforce Builder walkthrough covers Preview and activation.

Why can’t my agent find anything?

Salesforce’s troubleshooting article lists these checks. I’d go in this order:

1. Is the index ready? On the library’s search index in Data 360, the status should be Ready.
2. Are the articles published? Only published articles count. With Use Public Knowledge Articles on, only articles visible in the public knowledge base are returned.
3. Are the data categories visible? If you filter by category, check category visibility in profiles, permission sets and roles.
4. Can the agent user read Knowledge? By default the Agentforce Service Agent doesn’t have Read access to the Knowledge__kav object. Also check that the agent user has the AgentforceServiceAgentUserPsg permission set group.
5. Are your files mostly images? Only text is chunked. Structured, well-formatted documents work best.
6. Still nothing? Rebuild the search index. Its name starts with KA_ for Knowledge or FileUDMO_SI for files.

Can I manage data libraries with code?

Yes. The ADL API (Agentforce Data Library) in Connect REST API lets you create a Knowledge library, scope it to data categories, start indexing and check its status.

A few rules from the developer guide: you can’t update a library while indexing is running, and you can’t delete one that active agents still use.

Studying for the exam? The Agentforce Specialist study guide covers Data Library and grounding, and the certification study guides compare every cert.

Are you grounding on Knowledge or on files? Let me know!

Hope this helps!

Best,

Nick