Workflow ai.agent prompt gives completely different answer than when asking the same agent in the chat interface

Elastic Search
,
1

So I'm messing around with workflows and found out that the answers i get from that prompt are completely different and less expansive than when I copy-paste that prompt into the chat with the exact same agent. this is the step:

  - name: converse_with_agent

    type: ai.agent

    with:

      agent_id: "outsystems-log-assistant_1"

      message: |

        "

        An Elastic ML anomaly detection job has fired indicating abnormal SQL Server CPU utilization on a specific database server in the OutSystems production environment.

        can you analyze what could be causing this?

        "

when I ask it in the chat, I get a very useful expansive answer that dives very deep into the logs and finds stuff I didn't even think of which made me very happy. However when asked in the workflow step, the answer is very short with just some basic information that is almost completely redundant.

Am I missing some options for this workflow step or are there built-in limitations for the ai.agent type workflow step?

I tried to search the docs but couldn't find anything useful.

Thanks for the help!

2

Hello and welcome,

What is the version of your cluster?

Also, what is the model you used in the chat? It may be that the workflow is using a different model.

3

hey Tiemen, its Shahar from the workflows team.

there is no different default configuration, so its either:

  1. not the same agent

  2. different prompt (context/engine)

  3. different model (did you use the default LLM on the UI?

(if you want, you can join our workflows

channel on Elastic's OSS slack workspace, we tend to be very available there :slight_smile: )