Tool Calls

Tool Calls Observability

Track and analyze tool calls made by AI agents to understand their behavior and debug issues.

What are Tool Calls?

Tool calls are function invocations made by AI agents to interact with external systems, APIs, or perform specific actions.

Logging Tool Calls

{
  "timestamp": "2024-01-15T10:30:00Z",
  "agent_id": "agent-123",
  "tool_name": "search_database",
  "input": {"query": "user data"},
  "output": {"results": 42},
  "duration_ms": 150,
  "status": "success"
}

Key Metrics

• Call Frequency - How often each tool is called
• Success Rate - Percentage of successful tool calls
• Latency - Time taken for tool execution
• Error Types - Common failure modes

Best Practices

1. Log both input and output of tool calls
2. Include timing information
3. Track the agent context for each call
4. Set up alerts for high failure rates