Data Accuracy

The best way to validate an Agent's output is to compare it against the source data in NinjaCat's Data Cloud. Here's how:

1. Use Data Explorer – Filter down to the relevant data in Data Cloud and compare it with the Agent's response. Since this is the data the Agent has access to, the results should align.
2. Avoid comparing Agent outputs to reports or templates – The Template Editor and Reports in NinjaCat may not be an accurate reference. If there's a discrepancy between the Agent's output and a report, check Data Cloud first.
3. If Data Cloud results don't match the report – This could indicate an ingestion issue, not an Agent issue. If you suspect a data ingestion problem, please submit a ticket to [email protected].

Agent Output & Formatting

Fine-tuning an Agent's output takes time. After the initial setup, test the Agent by having a few conversations and reviewing its outputs. Ask yourself:

• Is the Agent presenting a table with all the columns I requested?
• Does the analysis include all expected details and recommendations?
• Is it pulling the correct data fields (e.g., Campaign Name, CTR calculations)?

If something isn't quite right, it's completely normal to go back into the Agent Builder and:

• Tweak the prompt → Test the response → Adjust again

This iterative process helps refine the Agent until you're confident in its performance. With each refinement, the Agent gets better at delivering the insights you need.

Error States & System Messages

The AI Agents chat interface occasionally surfaces system messages that aren't related to your prompt or your data. The patterns below cover what's most often behind them and what to try first. For anything that doesn't resolve quickly, escalate to NinjaCat Support — see guidance at the bottom of this section.

"Something went wrong" or a similar generic error

This is almost always a transient system issue, not a problem with how your Agent is configured.

• Refresh the page and retry the message.
• Wait a minute and try sending the message again.
• If the error appears in a long-running conversation, start a new conversation and try once more — see Chatting With Agents for context window guidance.
• If the error persists across multiple conversations, multiple Agents, or multiple users in your organization, escalate to NinjaCat Support.

No response, blank response, or the Agent appears stuck

Most often this is a context-window or query-complexity issue rather than a system error.

• Simplify the question and break it into smaller, more focused asks.
• Narrow the date range — for example, ask about the last 14 days instead of the last 90 days.
• Confirm the dataset connected to the Agent actually contains data for the period and accounts you're asking about.
• If you've been chatting for a while, start a new conversation to reset the context window.
• See Chatting With Agents for details on context windows and the 100,000-row Query Executor limit.

Session, connection, or loading errors

If the chat interface itself fails to load, disconnects, or repeatedly errors out before the Agent can respond, the issue is usually local to your browser or network.

• Check your internet connection.
• Hard refresh the page (Cmd/Ctrl + Shift + R).
• Try an incognito or private window to rule out browser extensions.
• Clear your browser cache and cookies for ninjacat.io.
• Try a different browser (Chrome, Firefox, Safari, Edge).
• If the issue persists across browsers and devices, escalate to NinjaCat Support.

When to escalate to NinjaCat Support

If the steps above don't resolve the issue — or if the same error is hitting multiple users in your organization — open a support ticket and include:

• What you were asking the Agent to do (the prompt or a description of the task).
• The exact error message or behavior you saw, and a screenshot if possible.
• Which Agent you were using, and which dataset(s) it was connected to.
• A rough timestamp (date and approximate time, with timezone) so we can trace it in our logs.
• Whether the issue is reproducible, intermittent, or hit only once.

For a full catalog of error messages you may see in AI Agents chat — including rate limit, provider overload, context window, credential, and file upload errors — see AI Agents Chat: Error Messages & Troubleshooting.