API tutorials

RobustDX API

Retrieve generated adversarial prompts and submit each target-model response with advResponse.

RobustDX uses the same trace-based pattern as BenchDX, but its endpoint paths and response field are different. Wait for adversarial prompt generation to finish before retrieving the prompt list.

1. Get adversarial prompts

GET/api/evaluation/all-adv-prompts

Returns all generated prompts for the selected RobustDX evaluation.

Location	Name	Type	Required
Query	`evaluationId`	number	Yes
Header	`Authorization`	Bearer PAT	Yes

Get adversarial promptscURL

curl -X GET \
  "https://trial.aidx.pro/api/evaluation/all-adv-prompts?evaluationId={evaluation_id}" \
  -H "Authorization: Bearer {personal_access_token}"

Example responseJSON

{
  "code": "req_01J...",
  "message": "Operation Succeed",
  "data": [
    {
      "traceId": "trace_id_1",
      "content": "adversarial_prompt_1"
    },
    {
      "traceId": "trace_id_2",
      "content": "adversarial_prompt_2"
    }
  ],
  "success": true
}

2. Run the target model

Send every adversarial content value to the target model. Keep the original traceId paired with the answer produced for that prompt.

3. Submit an adversarial response

POST/api/evaluation/adv-response

Records one target-model answer for a RobustDX trace.

Location	Name	Type	Required
Form	`evaluationId`	number	Yes
Form	`traceId`	string	Yes
Form	`advResponse`	string	Yes
Header	`Authorization`	Bearer PAT	Yes

The API accepts multipart/form-data and application/x-www-form-urlencoded parameters.

Submit adversarial responsecURL

curl -X POST \
  "https://trial.aidx.pro/api/evaluation/adv-response" \
  -H "Authorization: Bearer {personal_access_token}" \
  -F "evaluationId={evaluation_id}" \
  -F "traceId={trace_id}" \
  -F "advResponse={model_response}"

Example responseJSON

{
  "code": "req_01J...",
  "message": "Operation Succeed",
  "data": "Operation success",
  "success": true
}

4. Verify progress

Use Process Details to confirm each trace is evaluated. The evaluation can reach Completed only after every required test case is evaluated or explicitly handled.