Skip to main content

Agent Intelligence — LLM & Tool Behaviour

Access: Tenant Admins only

Agent Intelligence is part of the Analytics dashboard. It focuses on the internal activity of your agents — how many LLM calls they make, which tools they invoke, and where errors come from.

Analytics — Agent Performance and Tool Usage tables

Key Metrics to Watch

MetricWhere to Find ItWhat to Look For
Total LLM CallsRow 2 stat cardHigh count vs requests = multi-step reasoning or loops
LLM calls / response avgBelow Total LLM Calls cardAbove 5 calls/response = agent may be looping
Tool CallsRow 2 stat cardLow count for tool-enabled agents may mean hallucination
Tool Error RateRow 2 stat cardShould be 0% — any errors need investigation
Top ModelsTop lists panelConfirms which LLM model your agents are using
Top ToolsTop lists panelShows which tools are most relied upon

Daily LLM & Tool Activity Chart

The grouped bar chart compares LLM calls vs Tool calls per day.

Healthy pattern: Tool calls should track proportionally with LLM calls — if LLM calls spike but tool calls stay flat, the agent may be reasoning in circles without grounding its answers in real data.


Error Breakdown Chart

Shows errors grouped by type with percentage share:

Error TypeCommon Cause
AgentGuard BlockedAgentGuard blocked the prompt or response
429Rate limit hit on the LLM provider
UnknownUnhandled exceptions — check Traces for the stack trace

Warning Signs

PatternWhat It May IndicateAction
LLM calls / response > 5Agent is looping or over-reasoningReview system prompt, add conciseness instructions
Tool Error Rate > 0%Tool integration brokenCheck tool configuration and credentials
AgentGuard blocks dominatingAgentGuard blocking legitimate messagesReview AgentGuard settings in Content Safety
Error rate ≥ 5% on an agentSystematic failureSort Agent Performance table by Error Rate and check Traces

Common Use Cases

ScenarioWhat to Do
"My agent seems slow"Check Avg Latency card and sort Agent Performance by Avg Latency
"My agent isn't using its tools"Check Tool Calls count and Top Tools — if blank, tool config may be wrong
"Lots of AgentGuard blocks"Go to AgentGuard → Content Safety to review what's being blocked
"I want to compare two agents"Filter by each agent separately and compare LLM call counts and error rates