One conversation as a readable transcript: who said what, in order, plus the tool calls the agent made along the way. Deliberately not the full session card (`GetSessionHistory`): that one carries per-message metrics, model references and presigned audio links. Readers — the in-app agent above all — pull a dozen conversations at a time to judge how an agent behaves, and drown in the rest of it.

{
  "success": true,
  "data": {
    "session_id": "1ffd059c-17ea-40a8-8aef-70fd0307db82",
    "agent_id": "2b1e3b65-2c04-4fa2-a2d7-467901e98978",
    "agent_title": "string",
    "status": "Pending",
    "created_at": "2019-08-24T14:15:22Z",
    "duration_seconds": 0,
    "language_locale_code": "string",
    "error_message": "string",
    "test_eval_result": {
      "success": true,
      "reasons": [
        "string"
      ],
      "completion_reason": "StepsExhausted"
    },
    "turns": [
      {
        "role": "System",
        "kind": "Message",
        "text": "string",
        "at": "2019-08-24T14:15:22Z",
        "interrupted": true,
        "tool_name": "string",
        "tool_arguments_json": "string",
        "tool_output": "string",
        "tool_is_error": true
      }
    ]
  },
  "error": "string"
}