This short tutorial is go through how to start capturing traces from 3rd party AI systems and ingest them in the AI Control Tower.
For the purpose of this example, I will use a LangGraph customer support Agent.
In order for our Agent to send its traces to the ServiceNow AI Control Tower (AICT), we need a (Traceloop) API keys. Follow step 1 in the document below to do so:
https://support.servicenow.com/kb?id=kb_article_view&sysparm_article=KB3144255
In AICT, go to Rules and Templates, then Evaluation and then click on the External AI Systems tab. You will then see the option to create an API key.
One you have the API Key, we need it in our code. For simplicity, I have stored it in an environment file that can be retrieved with load_dotenv.
I can check that my environment properties are properly loaded using env_utils double_check:
In my Python code, I added the following code:
Document the resource_attributes with the relevant information to identify the impacted Configuration Item. sn.aict.system.id is the mandatory one. It is the sys_id of the impacted CI in ServiceNow.
You can find the sys_id in the url of the CI you are looking at.
Once we run our LangGraph agent, logs will now be fed to the instance:
If we then go to the AI Control Tiower and select insights -> Monitor -> Evaluated sessions, we will then see our traces.
We can then drill down on specific sessions:
The traces will also be visible from the actual AI Asset view (from the inventory).
Hope this helps
