Add Agent Control to the App

Change into the agent controls folder:

cd ~/workshop/healthcare-assistant/4-app-with-controls

If you’re using the standalone version of Splunk Agent Observability for the workshop, run the following command to create a Kubernetes config map, which the application will use to configure Agent Controls:

kubectl create configmap splunk-agent-control-config \
  --from-literal=SPLUNK_AO_API_URL="https://api.multitenant.sao.splunkcloud.com" \
  --from-literal=AGENT_CONTROL_URL="https://console.multitenant.sao.splunkcloud.com/api/agent-control" \
  --from-literal=AGENT_CONTROL_AGENT_NAME="agent-control-example" \
  --from-literal=AGENT_CONTROL_API_KEY_HEADER="Splunk-AO-API-Key" \
  --from-literal=AGENT_CONTROL_RUNTIME_AUTH_MODE="jwt" \
  --from-literal=AGENT_CONTROL_TARGET_TYPE="agent_stream"

Alternatively, if you’re using Splunk Agent Observability within Splunk Observability Cloud for this workshop, please use the following command instead:

kubectl create configmap splunk-agent-control-config \
  --from-literal=AGENT_CONTROL_URL="https://app.$REALM.observability.splunkcloud.com/api/agent-control" \
  --from-literal=AGENT_CONTROL_AGENT_NAME="agent-control-example" \
  --from-literal=AGENT_CONTROL_API_KEY_HEADER="Splunk-AO-API-Key" \
  --from-literal=AGENT_CONTROL_RUNTIME_AUTH_MODE="jwt" \
  --from-literal=AGENT_CONTROL_TARGET_TYPE="agent_stream"

Confirm requirements.txt includes the Agent Control SDK and evaluators:

agent-control-sdk[splunk-ao]>=7.10.0
agent-control-evaluators>=7.10.0

In agent.py, we’ve added the Agent Control imports at the end of the import section:

from agent_control import ControlSteerError, ControlViolationError, control

The controls stage uses these in three places (already wired up in this folder):

• The LLM call is wrapped with @control(step_name=LLM_STEP_NAME), where LLM_STEP_NAME = "Healthcare Assistant", so the model’s responses can be evaluated, blocked, or steered.
• Each tool is registered as a controllable step (get_patient_info, delete_patient_record, and a shared retrieval_step for search tools) via the helpers in helpers/agent_control_helpers.py.
• The agent enables control spans on the Splunk AO logger (splunk_ao_logger.enable_agent_control()) and registers the steps with init_agent_control(...) so the console knows which steps exist for this agent.