FruxonDocs
ApiAgent judge

Judge an output against evaluation metrics

Scores `output` (e.g., an agent's captured reply) against one or more tenant evaluation metrics using a judge LLM borrowed from the agent's deployed revision, and returns the per-metric breakdown plus a weighted aggregate in [0, 1]. Unlike `:evaluate`, this does not run the agent — it grades text that already exists, which is what the SDK sandbox test runner's semantic `reply_judge` assertion needs.

POST
/v1/tenants/{tenant}/agents/{agent}:judge

Authorization

Bearer
AuthorizationBearer <token>

JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.

In: header

Path Parameters

agent*string

The unique identifier of the agent that lends the judge provider.

tenant*string

The tenant identifier

Request Body

The output to judge, optional input, and the metric ids.

output?string|null

The text being judged (e.g., an agent's captured reply). Required.

input?string|null

The prompt that produced Fruxon.Model.AgentJudge.JudgeOutputRequest.Output, included for input-grounded grading. Optional.

metricIds?|null

Tenant Fruxon.Model.Evaluation.EvaluationMetric ids to score against. Must contain at least one id; weights are equal and normalized, so a single metric makes the aggregate equal to its score.

[key: string]?never

Response Body

curl -X POST "https://api.fruxon.com/v1/tenants/string/agents/string:judge" \  -H "Content-Type: application/json" \  -d '{}'
{
  "aggregate": 0,
  "metrics": [
    {
      "key": "string",
      "score": 0,
      "weight": 0,
      "rationale": "string"
    }
  ],
  "summary": "string",
  "providerId": "string",
  "model": "string"
}
{
  "type": "string",
  "title": "string",
  "status": 0,
  "detail": "string",
  "instance": "string",
  "property1": null,
  "property2": null
}
Empty
Empty
{
  "type": "string",
  "title": "string",
  "status": 0,
  "detail": "string",
  "instance": "string",
  "property1": null,
  "property2": null
}