Skip to main content

Telemetry & Logs

Telemetry & Logs provides comprehensive insights into your agents' performance and behavior. Access real-time metrics, execution traces, detailed logs, and LLM performance data to troubleshoot issues, optimize efficiency, and monitor activity across all agents.

Key Benefits​

Telemetry & Logs helps you:

  • Monitor Performance: Track key metrics like conversation count, average response time, and error rates
  • Analyze Execution: View detailed traces and step-by-step execution logs for each run
  • Debug Issues: Identify component failures, logic breakdowns, and performance bottlenecks
  • Optimize LLM Usage: Review LLM performance metrics, model selection, and Time To First Byte (TTFB) data
  • Track Component Health: Analyze error rates and latency for individual components
  • Audit Activity: Access up to 30 days of historical data for all agents
INFO

Data is retained for 30 days across all plan tiers, giving you complete visibility into your agent's recent activity and performance trends.

Accessing Telemetry & Logs​

From the Sidebar​

Navigate to Telemetry & Logs in the main sidebar to view aggregated data across all agents. From here, you can:

  • Filter by specific agents using the agent dropdown
  • Adjust the time range for analysis
  • Explore different data views (Overview, Traces, Logs, LLM Performance, Components)

From Agent Settings​

Click on an agent card from the Agent Dashboard to open Agent Settings, then navigate to the Telemetry & Logs tab. This automatically filters all data to show only that agent's executions.

Dashboard Tabs​

Overview​

Get a high-level summary of agent performance with key metrics:

  • Conversations: Total number of agent interactions during the selected period
  • Average Conversation Time: Average duration of each agent execution
  • Skill Runs (Component Executions): Total number of component/skill invocations
  • Error Rate: Percentage of failed executions
  • Component Calls: Total count of individual component calls
  • Activity Over Time Chart: Visual trend of conversations, skill runs, and errors across the past 30 days
Tip

Use the Overview tab to identify trends and anomalies. A sudden spike in error rate may indicate a recent deployment issue or external service outage.

Traces​

View detailed execution traces for individual agent runs:

  • Trace ID: Unique identifier for each execution
  • Root: The primary entity executed (e.g., Agent.Skill)
  • Agent: The endpoint or agent that triggered the trace
  • Duration: Total execution time
  • Spans: Number of distinct operations within the trace
  • Errors: Count of errors encountered during execution

Click on any trace row to inspect the complete execution flow and step-by-step component interactions.

Logs​

Access structured logs for detailed event tracking:

  • Timestamp: When the event occurred
  • Severity: Log level (INFO, WARN, ERROR, etc.)
  • Message: Description of the logged event
  • Agent: Associated agent or endpoint
  • Trace ID: Link to the related trace for context

Filter and search logs to quickly locate specific events or error messages.

LLM Performance​

Monitor Large Language Model (LLM) usage and performance:

  • Total LLM Calls: Aggregate count of LLM API calls
  • Average TTFB (Time To First Byte): Average response latency from the LLM
  • TTFB by Model: Breakdown of response times across different LLM models
  • Model Usage: Which LLM models are being used and how frequently
  • Response Time Metrics: Additional timing and performance data

Use this tab to compare LLM providers, identify slow models, and optimize model selection.

Optimize LLM Costs

Monitor TTFB and model usage to identify opportunities for model switching or prompt optimization.

Components​

Analyze individual component performance and reliability:

  • Errors per Component: Pie chart showing error distribution across components
  • Latency per Component (Top 10): Bar chart displaying response times for the most-used components
  • Component Calls: Frequency of invocations for each component

Identify problematic or slow-running components and prioritize optimization efforts.

Detailed Log Inspection​

View a Specific Run​

Click on any trace or log entry to open the detailed execution view:

  • Component Name: Name of the executed component
  • Component Type: Classification (e.g., API call, GenAI, code execution)
  • Input Payload: Data sent to the component
  • Output Payload: Results returned by the component
  • Execution Status: Success, failure, or warning indicators
  • Timing: Duration of component execution

Call Tree Visualization​

The Call Tree provides a visual map of your agent's execution flow:

  1. Open a trace from the Traces tab
  2. Click Show Call Tree to visualize the execution path
  3. A hierarchical tree displays each component, data transformation, and decision point
  4. Follow the flow to identify silent failures or skipped branches
Tip

Use the Call Tree to trace data flow through complex workflows and identify where logic diverges from expectations.

Time Range Selection​

By default, Telemetry & Logs displays data from the past 30 days. You can adjust the time range to focus on:

  • Recent activity (last 24 hours, 7 days)
  • Historical analysis (full 30-day window)
  • Specific incidents or events

Filter by Agent​

When viewing Telemetry & Logs from the sidebar, use the Agent dropdown to filter:

  • All Agents: View aggregated data across your entire workspace
  • Specific Agent: Focus on a single agent's telemetry

Search and Sort​

  • Sort logs by timestamp, severity, agent, or trace ID
  • Search logs by message content or trace ID
  • Filter traces by error status or duration ranges

Common Use Cases​

Debugging Agent Failures​

  1. Go to the Overview tab and note spikes in the Error Rate
  2. Navigate to the Traces tab and filter for failed runs
  3. Click on a failed trace to view the Call Tree
  4. Inspect component inputs and outputs to identify the failure point
  5. Check the Logs tab for error messages and stack traces

Optimizing Performance​

  1. Review Average Conversation Time on the Overview tab
  2. Identify slow components using the Components tab
  3. Compare LLM performance across models in the LLM Performance tab
  4. Adjust model selection, prompt engineering, or component logic based on findings

Monitoring LLM Costs​

  1. Check Total LLM Calls and TTFB by Model in the LLM Performance tab
  2. Identify underperforming or expensive models
  3. Consider switching to faster or more cost-effective alternatives

Troubleshooting Specific Issues​

  1. Search the Logs tab for error messages or component names
  2. Click the associated Trace ID to view full execution context
  3. Download the JSON log for detailed analysis or sharing with support
INFO

All workspace members can access Telemetry & Logs with data available for the past 30 days, regardless of plan tier.

What to Explore Next​