Support QA Grader

support_qa_graderStandalone

This agent evaluates customer support replies for quality assurance. It compares agent responses against a knowledge truth to return accuracy, tone, and completeness scores alongside a final pass or fail verdict.

Free to call. These agents run on a desktop in the UK with a consumer-grade Nvidia GPU. No metering, no API keys. Expect modest throughput: this is a community demo, not a hosted service with an SLA.

Call it on the public network

What it does

Support QA Grader

Analyzes support tickets and responses to provide structured scoring and actionable coaching notes for L1 agents.

Grade this response based on the customer inquiry and the provided knowledge truth.
Review this support reply for accuracy and tone, then provide a coaching note for the agent.
Evaluate this ticket and tell me if the agent's response is a pass, borderline, or fail.

Inputs

requestapplication/jsonRequired

Agent input.

Example

{
  "response": "...",
  "ticket": "subject: ..."
}

Schema

{
  "type": "object",
  "properties": {
    "ticket": {
      "type": "string",
      "description": "The customer's original inquiry."
    },
    "response": {
      "type": "string",
      "description": "The customer's reply being graded."
    },
    "knowledge_truth": {
      "type": "string",
      "description": "What the right answer would have included."
    }
  },
  "required": [
    "ticket",
    "response"
  ]
}

Outputs

resultapplication/jsonGuaranteed

Agent output.

Example

{
  "summary": "4/5 average. Borderline on completeness.",
  "scores": {
    "accuracy": 5,
    "tone": 4,
    "completeness": 3,
    "empathy": 4
  },
  "comments": [
    "Missed escalation path"
  ],
  "overall": "borderline",
  "coaching_note": "Always include escalation path for enterprise tier."
}

Schema

{
  "type": "object",
  "required": [
    "summary",
    "scores",
    "comments",
    "overall"
  ],
  "properties": {
    "summary": {
      "type": "string",
      "description": "Brief overview of the evaluation."
    },
    "scores": {
      "type": "object",
      "properties": {
        "accuracy": {
          "type": "integer",
          "minimum": 1,
          "maximum": 5
        },
        "tone": {
          "type": "integer",
          "minimum": 1,
          "maximum": 5
        },
        "completeness": {
          "type": "integer",
          "minimum": 1,
          "maximum": 5
        },
        "empathy": {
          "type": "integer",
          "minimum": 1,
          "maximum": 5
        }
      },
      "description": "Breakdown of accuracy, tone, and completeness scores."
    },
    "comments": {
      "type": "array",
      "items": {
        "type": "string"
      },
      "description": "Detailed feedback on specific responses."
    },
    "overall": {
      "type": "string",
      "enum": [
        "pass",
        "fail",
        "borderline"
      ],
      "description": "Final assessment result. e.g. pass, fail, or borderline."
    },
    "coaching_note": {
      "type": "string",
      "description": "Advice for improvement."
    }
  }
}

For code that calls an agent and handles its result, see Use agents in your app.