The Knowledge Tab

The Knowledge Tab of the Agent Builder is where you give your Agent access to the information it needs to perform its tasks. This information comes in the form of Datasets (or Views of those Datasets) from the Data Cloud, and/or files of information.

📘

An Agent cannot analyze Google Ads data if it only has access to Facebook Ads data—or no data at all.

How to Assign Datasets in the Knowledge Tab

If an Agent's goal involves analyzing your advertising data, it must be connected to the relevant datasets to generate accurate insights. Without assigned data, the Agent can only rely on general AI knowledge and will not be able to interact with your business-specific information.

1. Select a Dataset (or View) – Choose the specific Dataset or View that the Agent should access and interact with.
2. Name Your Dataset – Assign a friendly name that makes it easier for the Agent to reference in responses.
3. Add a Description (Optional) – Provide context to help the Agent better understand the Dataset.
4. Review Available Columns – Preview the columns within the Dataset that the Agent can interact with.

📘

If the "Data Cloud Access" permission has been enabled for your user, then each Dataset or View includes a "Preview Dataset" icon. Selecting this icon will open the Data Explorer, where you can examine the dataset’s structure and contents in more detail.

If your user profile does not have access to Data Cloud, then you will not see that Preview Dataset icon, but you still have the ability to manage the Datasets assigned here.

Optimizing Agent Performance with Data

The more datasets an Agent has access to, the more specific its prompt needs to be to ensure accurate results. To improve accuracy of the Agent output, reducing ambiguity is necessary. The best way to reduce ambiguity:

• Use Data Explorer to filter the data down to only what is necessary and save that as a View. This could be removing unnecessary columns, filtering the data by a particular Network or Campaign type that the Agent needs. Then, assign the View to the Agent instead of the entire Dataset.

📘

By carefully selecting and refining the data assigned in the Knowledge Tab, you ensure that your Agent delivers the most precise and relevant insights for your needs.

How to Upload Files in the Knowledge Tab

1. Select "Add File"
2. Locate the file(s) you wish to upload (multiple can be selected at one time)
   1. Supported files in the Agent's Knowledge: PDFs and all text format files (including but not limited to: .txt, .csv, .md, .json, .htm, .html, .py)
      1. Note: image files are only supported directly in a conversation w/ an agent, not yet supported under Knowledge.
   2. There is a 30 MB limit on individual files that can be added
   3. You may select more than one file at a time, however there is a 50 MB limit per upload request. It is recommended to just add files one a time to get around this 'per request' limit.
   4. There is no limit to the number of files that can be uploaded
3. Once a file is uploaded to the Agent, the system auto-generates a description of that file which can be used by the Agent to help it understand when to pull that file and its context into the conversation / the context window. To view this description, select the More Actions icon on the individual file listing, and select Edit.
4. A popup will surface displaying the AI generated version of the file description. Review this to ensure its explaining the contents appropriately so that the agent knows when to use it (you may manually edit this description). Generally, make sure it includes the below, but does not need to be a very long description:
   1. What the file contains
   2. When the agent should use the file for a conversation
   3. Any important context or limitations.

The Agent will be able to reference the text of those files for any future conversations, using both the filename and the file description to determine whether the content is necessary for the conversation. Just the text of those uploaded files will be understood by the Agent. The Agent won't understand any images or charts or blocks of color that are apart of the files.

Note: the auto-generation of the file description feature was released on 1/29/26. For any files that were added to an agent prior to this date, there will not be an auto generated description available. Please add one manually.

📘

The text context of these files gets inserted into the Context Window of the conversation (only when its needed for the conversation), which does have a token limit. Once that limit is hit, the conversation will end in error. Do keep this in mind when considering the amount of files to upload to one Agent's knowledge, especially for a very lengthy conversation. Learn more about Context Windows here.