diff --git a/ui/litellm-dashboard/src/components/AgentMonitoring.tsx b/ui/litellm-dashboard/src/components/AgentMonitoring.tsx index 26186ba1574..4215fb3fbbc 100644 --- a/ui/litellm-dashboard/src/components/AgentMonitoring.tsx +++ b/ui/litellm-dashboard/src/components/AgentMonitoring.tsx @@ -198,9 +198,9 @@ function AgentDetail({ agentId, onBack, accessToken, token, userRole, userID, al
High drift detected — Claims Processing Agent
+Drift alert — Claims Processing Agent
- The semantic similarity between this agent's description ("answer member questions about insurance claims") and the tools it is calling (exec, write, fetch_claims) is very low. The gateway embeds both on every run and computes cosine similarity — this agent is scoring consistently near zero. + This agent is described as "answer member questions about insurance claims" but is calling exec, write, and fetch_claims. On every request, the gateway embeds the agent description and the tools called, then computes cosine similarity between them. This agent is scoring 0.08 — far outside the 0.82–0.95 range we see for healthy claims agents.