Safety & DetectionDeepfake DetectionDetect Intelligence

Detect Intelligence

Ask natural-language questions about completed deepfake detection reports using the detection’s full context.

This is different from the standalone Intelligence feature, which analyzes raw media for speaker details, transcription, and emotion. Detect Intelligence asks questions about already-completed detection reports.

Request & Response Schemas

1interface AskDetectQuestionRequest {
2  query: string; // natural-language question about the detection
3}
4
5interface DetectQuestionItem {
6  uuid: string;
7  detect_uuid: string;
8  query: string;
9  answer: string | null;       // null until status is "completed"
10  status: "pending" | "processing" | "completed" | "failed";
11  error_message: string | null;
12  created_at: string;          // ISO 8601
13  updated_at: string;          // ISO 8601
14}

Async Workflow

  1. Ensure the detection status is completed.
  2. Submit a question via the Ask a Question endpoint.
  3. The API returns 202 Accepted with the question UUID and status: "pending".
  4. Poll the Get Answer endpoint until status is "completed" or "failed".
  5. Read the answer field from the completed response.

Example Questions

CategoryExample Query
Summary”Summarize the detection results in plain language.”
Confidence”How confident is the model that this audio is fake?”
Comparison”How do the audio and video detection results differ?”

Polling Best Practices

  • Use exponential backoff starting at 1 second, doubling up to a maximum of 10 seconds.
  • Most answers complete within 5–15 seconds.
  • Check the status field — stop polling once it reaches completed or failed.