Knowledge
How to Troubleshoot Knowledge in KriyaOS
If your agent is not answering the way you expected, the problem is often not the agent itself.
Very often, the issue is that the knowledge is missing, not ready yet, or different from what you thought you added.
Most Common Root Cause
If an agent sounds too generic, the first thing to check is usually the knowledge, not the model.
Start With Basic Knowledge Troubleshooting Checks
Before changing prompts or settings, ask three simple questions:
- Did I actually add the knowledge?
- Is that knowledge ready to use yet?
- Am I expecting the agent to answer something the knowledge does not really cover?
These three checks solve many early issues.
Common Reason 1: No Knowledge Was Added
Sometimes the agent is working exactly as configured, but it does not have the business information you expected.
If no useful knowledge was added, the agent may still answer, but it will not answer with your specific business context.
Common Reason 2: Knowledge Is Not Ready Yet
If you uploaded files recently, they may still be processing.
That means:
- the upload may have worked
- but the agent may not be able to use it yet
If the behavior feels too generic right after upload, readiness is one of the first things to consider.
Common Reason 3: The Wrong Expectations
Sometimes users expect the agent to know something that was never actually included in the knowledge.
For example:
- the document did not contain the answer clearly
- the uploaded information was too broad or too vague
- the question being tested is outside the actual content
In those cases, the agent is not necessarily broken. It may simply not have the information you expected it to use.
How To Debug Agent Behavior
A simple way to debug is:
Step 1: Check The Source
Look at the knowledge you added and ask:
- does it really contain the information I want the agent to use?
- is it clear enough?
Step 2: Check Readiness
If the knowledge was added recently, allow time for processing.
Step 3: Test A Simple Question
Start with a direct question that should be easy to answer from the knowledge.
If even the obvious question does not work, the problem is easier to spot than if you start with a complex one.
Step 4: Separate Prompt Problems From Knowledge Problems
If the agent sounds wrong, the prompt may need work.
If the agent lacks the right facts, the knowledge may need work.
Often these are different problems.
What To Improve First
If the issue seems knowledge-related, improve in this order:
- make sure the right knowledge exists
- make sure it is ready
- test clear, direct questions
- only then adjust the prompt if needed
A Good Rule Of Thumb
If an agent gives vague or generic answers, do not assume it is failing.
First check whether:
- it has the right knowledge
- that knowledge is ready
- your test question matches the information you actually provided