The AI Agent tab shows the performance and consumption of the project's AI agents and their tools. Do not use these figures for billing. They are not real time but are refreshed at intervals.
| Widget | What it shows |
|---|---|
| Total Agent Requests | The number of requests the AI agents processed. |
| Error Rate | The percentage of agent requests that ended in an error. |
| Average Time to First Token | How quickly the first token of a response is produced after a request arrives. |
| Total Tool Calls | The number of tool calls the AI agents made. |
| Total Cost, Average Cost per Session, Cost per Minute | The estimated cost of the AI agents' model usage. |
| Input Cache Savings | Cached against uncached input, and how many tokens caching saved. |
| Tool Performance (Run Count & Latency) | Run count per tool as bars and average latency as a line. Use Show top 10 results for to rank the tools by Run Count or by Average Latency. |
| Agent Request Distribution | The requests handled by each AI agent. |
| Tool Call Distribution | Each tool type's share of the tool calls, for example Workflow, API Integration, MCP or Knowledge Retrieval. |
| Input Token Consumption Trend, Output Token Consumption Trend | Token usage over time. Guardrail tokens are not included. |
| Time to First Token Trend | Average, 95th percentile and 99th percentile time to first token, over time. |
| AI Agent Performance Overview | A table with one row per agent: Total Calls, Input Tokens, Cached Input, Output Tokens, Total Tokens and Cost. |
| Tool Performance Overview | A table with one row per tool: AI Agent, Run Count, Average Latency and Error Rate. Sort by any column to find slow tools or tools that fail often. |
The cost cards depend on the same tenant feature as Cost Analysis: if that tab is missing, these cards are too.
Limits
| Item | Value |
|---|---|
| Tool Performance (Run Count & Latency) | The top 10 tools |