Docs

Quickstart

Send the transcript of a call your voice agent handled, and read back a QA result in which every potential violation quotes the transcript line it is based on. Results are QA signals for human review.

What this is, and is not

This is post-call Voice QA: it evaluates calls that have already happened. It is not a pre-deployment simulation or test harness for your agent.
Three allowances are counted separately each month: QA calls, minutes of audio sent to speech-to-text, and optional AI assessments. A call that arrives as a transcript uses no audio minutes.
Speaker labels decide what can be established. Without them, checks that depend on who spoke can only be a review, never a pass or a violation.
“Review recommended” is not a proven violation: it means a person should look, usually because something could not be confirmed.
A potential violation always carries the quote it is based on, with its timestamp when your transcript has times; some reviews (for example a call that looks incomplete) have none.

1. Create an organization and an API key

Sign up, then open Settings & API → API keys (organization owners and admins). The key is shown once; only a one-way hash is stored. Keep it on your server. Export it for the examples below:

export GOLDNOVA_API_KEY="gnk_…"

2. Send a call

externalCallId is your own identifier for the call. Sending the same call again is safe: it is stored once and answered with "duplicate": true.

Request

curl -X POST https://goldnovaai.com/api/v1/calls \
  -H "Authorization: Bearer $GOLDNOVA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "externalCallId": "call-1042",
    "agent": { "id": "support-bot-1", "name": "Support bot" },
    "transcript": [
      { "speaker": "agent", "text": "Hello, this call is recorded for quality purposes.", "startTime": 0, "endTime": 4 },
      { "speaker": "customer", "text": "Okay.", "startTime": 4, "endTime": 5 },
      { "speaker": "agent", "text": "We guarantee a full refund.", "startTime": 41.2, "endTime": 44.0 }
    ]
  }'

Response (201; 200 with duplicate: true when the identical call was already stored)

{ "id": "6f1c0a52-8d1e-4c43-9a7b-2f6a3d9e1b10", "externalCallId": "call-1042", "status": "ready", "duplicate": false }

3. Read the result

Evaluation runs after the response, usually within seconds. status moves received → ready → evaluating → evaluated (or failed). Poll the call until it is evaluated.

Request

curl https://goldnovaai.com/api/v1/calls/6f1c0a52-8d1e-4c43-9a7b-2f6a3d9e1b10 \
  -H "Authorization: Bearer $GOLDNOVA_API_KEY"

Response

{
  "id": "6f1c0a52-8d1e-4c43-9a7b-2f6a3d9e1b10",
  "externalCallId": "call-1042",
  "status": "evaluated",
  "transcriptSource": "provided",
  "transcriptQuality": null,
  "failureReason": null,
  "evaluation": {
    "result": "violation",
    "resultLabel": "Potential violation",
    "qaScore": 62.5,
    "policyVersion": 3,
    "evaluatedAt": "2026-10-01T09:31:02.114Z"
  },
  "findings": [
    {
      "rule": "Prohibited claims",
      "ruleType": "prohibited_claim",
      "severity": "high",
      "status": "violation",
      "explanation": "The agent used a prohibited phrase.",
      "evidence": [{ "quote": "We guarantee a full refund.", "startTime": 41.2, "endTime": 44.0 }]
    }
  ],
  "notice": "QA signal for human review. Findings are automated indications, not legal or compliance determinations."
}

How to read a result explains the three result values, the score, severities and why a finding can only be shown with a verbatim quote.

Limits at a glance

Request body	1.5 MB
Segments per call	5,000, each up to 4,000 characters
`transcriptText`	up to 400,000 characters
Metadata	up to 50 keys and 16 KB
Call duration	up to 24 hours
Rate limit	120 requests per minute per key (`429` with `Retry-After`)
Allowances	QA calls, audio transcription minutes and AI assessments are separate monthly allowances (see pricing); `402 quota_exceeded` when the call allowance is used up