Execution Runs & Step Traces

Deep execution tracing for multi-step agent actions, tool payloads, latency watermarks, and loop abortions.

MongoDB `runs` Collection

Recent Agent Runs

Select a run to inspect its step waterfall

Step Waterfall: run_active_44

running

Agent: CEO • User: usr_giufeeRzEkoyj35vbtiEb • Cost: $0.0553

1search_web
820ms•1540 tok•$0.0046
Input Payload:
{
  "query": "Latest AI benchmark trends Q3 2026"
}
Output Result:
{
  "resultsCount": 8,
  "topDomain": "arxiv.org"
}
2steel_browser_navigate
2400ms•3100 tok•$0.0093
Input Payload:
{
  "url": "https://paperswithcode.com/sota"
}
Output Result:
{
  "domNodes": 1420,
  "pageTitle": "SOTA Leaderboard"
}
3e2b_python_interpreter
3100ms•5050 tok•$0.0151
Input Payload:
{
  "code": "import pandas as pd\ndf = pd.read_json('results.json')\nprint(df.describe())"
}
Output Result:
{
  "stdout": "Mean accuracy: 94.2%\nStd: 1.8%"
}
4format_report_markdown
1200ms•9800 tok•$0.0262
Input Payload:
{
  "sections": 4
}
Output Result:
{
  "status": "synthesizing"
}