Adding Other Knowledge

Adding Knowledge Files at the Agent Level

The Knowledge Tab of the Agent Builder lets you upload files directly to an individual agent. These files are available to that agent on every request — regardless of which account is being discussed.

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 once)
   • All file formats are supported — there are no restrictions on file type, except:
     • Image files are only supported directly in conversation with an Agent — not in Agent Knowledge.
   • 30 MB limit per individual file
   • 50 MB limit per upload request; recommended to add files one at a time to work around this limit
   • No limit on the total number of files that can be uploaded
3. Once uploaded, the system auto-generates a file description the Agent uses to determine when to reference the file. To view/edit: select the More Actions icon → Edit.
4. The popup shows the AI-generated description. Ensure it covers:
   • What the file contains
   • When the agent should use the file
   • Any important context or limitations

The Agent references the text content of uploaded files using both the filename and description to determine relevance. Images, charts, and color blocks within files are not understood by the Agent.

Note: Auto-generation of file descriptions was released on 1/29/26. Files added before this date will not have an auto-generated description — please add one manually.

📘

The text content of files gets inserted into the Context Window (only when needed), which has a token limit. Keep this in mind when uploading many or lengthy files to a single Agent. Learn more about Context Windows here.

Bob-Generated Knowledge Files

In addition to uploading files manually, Agent Builder Bob can create Knowledge Files for your Agent automatically during the agent-building process.

When Bob creates a file, it appears in the Knowledge tab alongside your uploaded files. You can distinguish between the two:

• Bob-generated files — Bob can create, edit, and delete these files. They are marked as Bob-generated in the file listing.
• User-uploaded files — Files you upload manually are protected. Bob can read them to understand the full context of your Agent's knowledge, but he cannot edit or delete them.

This distinction ensures that your manually curated content remains intact while still giving Bob the flexibility to manage the files he creates.

📘

For more details on how Bob creates and manages Knowledge Files, see Agent Builder Bob — Manage Knowledge Files.

Adding Knowledge at the Account Level

Agent Preferences is now available to all NinjaCat organizations. It gives you a dedicated space on the Account Command Center to attach account-specific context — instructions and Knowledge Files — directly to an individual account.

Why it matters

As you scale AI Agents across many client accounts, loading every client's context into the agent itself becomes unmanageable. Agent Preferences solves this: context lives outside the agent, attached to each account, and is pulled in automatically only when that account is relevant to the conversation. No re-uploading, no bloated agent configurations, no risk of one client's context bleeding into another's work.

What you can configure

Inside Agent Preferences, eligible users can set two things at the account level:

• Agent Instructions — A free-form text field for guidance that should always travel with the account. Examples:
  • Tone and voice ("formal", "casual", "use the client's terminology")
  • Reporting cadence and format preferences
  • Escalation rules and stakeholder preferences
• Knowledge Files — Account-specific files the agent can reference when working on this account. Examples:
  • Brand and style guidelines
  • SLA and billing rules
  • Annual advertising goals and KPIs the customer actually cares about
  • Any other context that's specific to this client

File limits

• Maximum 30 MB per Knowledge File
• Images are not supported

How it works

When a user @-mentions an account in an AI Agent conversation, the agent:

1. Checks that account's Agent Preferences.
2. Picks up any stored agent instructions and Knowledge Files.
3. Appends them to its working context for that request.

The result: the agent has everything it needs to work on that account — without any extra setup on your end.

Permissions

Access respects your existing account permissions. The agent only surfaces Agent Preferences for accounts the user is already permitted to view. There is no cross-contamination between clients.

Agent-level vs. account-level Knowledge Files

Use agent-level Knowledge Files for knowledge that should always be available to an agent. Use account-level Knowledge Files for client-specific context that should travel with the account.