🔬 Agent Autopsy

Framework-agnostic post-mortems for LLM agents. Paste an execution trace and get the real failure-pattern analysis — infinite loops, token storms, goal drift, error cascades — plus an execution flame graph.

1 · Load an example trace

2 · Or paste your own trace JSON

Agent trace (JSON)

{
  "name": "flight-booking-agent",
  "steps": [
    {
      "type": "DECISION",
      "description": "Goal: book the cheapest flight NYC -> London"
    },
    {
      "type": "LLM_CALL",
      "model": "gpt-4o",
      "prompt": "What should I do to find a flight from NYC to London?",
      "response": "I should search the flights API.",
      "latency_ms": 820,
      "tokens": 140
    },
    {
      "type": "LLM_CALL",
      "model": "gpt-4o",
      "prompt": "What should I do to find a flight from NYC to London?",
      "response": "I should search the flights API.",
      "latency_ms": 790,
      "tokens": 140
    },
    {
      "type": "LLM_CALL",
      "model": "gpt-4o",
      "prompt": "What should I do to find a flight from NYC to London?",
      "response": "I should search the flights API.",
      "latency_ms": 810,
      "tokens": 140
    },
    {
      "type": "LLM_CALL",
      "model": "gpt-4o",
      "prompt": "What should I do to find a flight from NYC to London?",
      "response": "I should search the flights API.",
      "latency_ms": 805,
      "tokens": 140
    },
    {
      "type": "ERROR",
      "error": "TimeoutError: flights API timed out (attempt 1)"
    },
    {
      "type": "ERROR",
      "error": "TimeoutError: flights API timed out (attempt 2)"
    },
    {
      "type": "ERROR",
      "error": "TimeoutError: flights API timed out (attempt 3)"
    }
  ]
}

Trace schema: { "name": str, "steps": [ { "type": "LLM_CALL|TOOL_CALL|DECISION|MEMORY_READ|ERROR", ... } ] }. LLM steps accept model, prompt, response, latency_ms, tokens, context_window.

3 · Autopsy report

Execution flame graph