Step 1: Upload & Process Traces
Upload traces to automatically run discovery, clustering, and rule extraction
Trace Format Reference
Your trace data must follow this JSON structure for proper processing
{
"traces": [
{
"trace_id": "string", // Unique identifier for the trace
"timestamp": "2024-01-01T00:00:00Z", // ISO 8601 datetime
"tenant": "string", // Tenant/organization identifier
"agent_id": "string", // Agent identifier
"agent_version": "string", // Version of the agent
"input": {}, // Full prompt + structured fields (object)
"candidate_actions": ["action1", "action2"], // List of possible actions
"selected_action": "action1", // The action that was selected
"pre_decision_signals": {}, // Optional: signals before decision (object)
"tool_calls": [ // List of tool calls made
{
"tool_name": "string", // Name of the tool
"inputs": {}, // Tool input parameters (object)
"outputs": {}, // Tool output results (object, optional)
"error_code": "string", // Error code if failed (optional)
"latency_ms": 123.45 // Execution time in milliseconds
}
],
"execution_metadata": {}, // Metadata: latency, cost, success/failure (object)
"outcome": "string", // User-facing outcome description
"context_snapshot": {} // Context at execution time (object)
}
]
}Required Fields:
trace_id- Unique identifiertimestamp- ISO 8601 datetimetenant- Organization IDagent_id- Agent identifieragent_version- Version stringinput- Input data objectcandidate_actions- Action listselected_action- Chosen action
Optional Fields:
pre_decision_signals- Decision contextoutputs- Tool outputs (in tool_calls)error_code- Error info (in tool_calls)
Notes:
- You can upload an array directly or wrap it in a
traceskey - All object fields accept any valid JSON structure
- Timestamps must be in ISO 8601 format
Upload Trace Data
Select a JSON file or paste JSON data directly. The system will automatically process and cluster the traces.