Skip to content

Vega Agent / Video Q&A โ€‹

Updated

Endpoint: https://smartmedia.vod-qcloud.com/aguiAuthentication: VOD ApiToken / Bearer

Retrieves videos ingested into the media knowledge base and answers questions about video content. Use forwardedProps.database to specify the knowledge base to query.

Basic Information โ€‹

ItemValue
ServiceVOD Media AI Agent
Request pathPOST /agui/chat
Scenariovideo-qa
InputQuestion, knowledge base name
PrerequisitesKnowledge base created and video ingestion completed
OutputRetrieval / tool execution events and answer text

Before You Call โ€‹

  1. Create a video knowledge base in the intelligent media feature of VOD.
  2. Ingest the videos you want to query and confirm that analysis is complete.
  3. Specify the knowledge base name with forwardedProps.database.

Media ingestion and question requests are separate steps. Simply writing a video URL in your question does not ingest that video into the knowledge base.

Request Parameters โ€‹

ParameterRequiredTypeDescription
forwardedProps.scenario_nameRequiredStringvideo-qa
forwardedProps.databaseOptionalStringKnowledge base name. Default value default
forwardedProps.modelOptionalStringVega Agent model ID
messages.roleRequiredStringQuestion input is user
messages.contentRequiredStringRetrieval or question answering request

Conversation ID, execution ID, and authentication follow API Reference.

Request Example โ€‹

json
{
  "threadId": "thread-001",
  "runId": "run-001",
  "messages": [
    {
      "role": "user",
      "content": "Find and summarize what the product demo videos say about battery life. Include the relevant videos and segments."
    }
  ],
  "forwardedProps": {
    "model": "wand-vega-agent-1.0-pro",
    "scenario_name": "video-qa",
    "approval_mode": "never",
    "database": "product-videos"
  }
}

Example Questions โ€‹

Use CaseExample Question
Scene searchFind a video that explains the product installation process.
Content checkSummarize the battery life described in the product demo video.
Information extractionList the safety inspection steps shown in the training video in order.

Specifying the search scope and the content to check in detail makes it easier to select the media needed for the answer.

Response Handling โ€‹

Calls to the retrieval tool and their results are delivered as TOOL_CALL_* events. The answer is displayed by concatenating TEXT_MESSAGE_CONTENT.delta by messageId. The field structure of events follows Response events.

Continuing a Conversation โ€‹

Follow-up questions use the same threadId and knowledge base name and specify a new runId. Previous conversations are retrieved with /agui/history, and in-progress requests are canceled with /agui/cancel.

Built with VitePress ยท Deployed on Tencent EdgeOne Pages