Vision AI Agents — Search Integration Guide

Vision AI Agents enables developers to retrieve insights from analyzed video content using powerful search APIs.

Search APIs allow developers to query indexed video intelligence signals using both structured filters and semantic search.

Search results return scene-level intelligence including timestamps, descriptions, and relevance scores.

Search Architecture

After videos are ingested and analyzed, Vision AI Agents stores intelligence signals in two formats:

Structured metadata indexes
Vector embeddings

These indexes power both traditional filtering and semantic search.

The search pipeline works as follows:

Video Ingest
      ↓
Video Intelligence Analysis
      ↓
Metadata Extraction
      ↓
Vector Embedding Generation
      ↓
Index Storage
      ↓
Search APIs
      ↓
Application Results

Developers interact with this pipeline through the Search API.

Search Endpoint

Endpoint

POST /api/search/query

This endpoint retrieves scene-level intelligence signals from the indexed video library.

Search Request Example

POST /api/search/query
Content-Type: application/json
Authorization: Bearer API_KEY

{
  "query": "high emotional engagement scene",
  "filters": {
    "genre": "drama"
  },
  "limit": 10
}

Search Parameters

Parameter	Type	Required	Description	Example
query	string	Yes	Semantic search query describing the type of scene or concept to retrieve	"high emotional engagement scene"
filters	object	No	Structured metadata filters used to narrow search results	{ "genre": "drama" }
limit	integer	No	Maximum number of search results returned	10

Filters allow developers to narrow search results using structured metadata.

Search Response Example

{
  "results": [
    {
      "video_id": "vid_8932jfks92",
      "timestamp": "00:02:34",
      "description": "Scene showing strong emotional engagement",
      "score": 0.93
    },
    {
      "video_id": "vid_8932jfks92",
      "timestamp": "00:05:12",
      "description": "Dialogue moment with rising tension",
      "score": 0.88
    }
  ]
}

Search Response Fields

Field	Type	Description
video_id	string	Unique identifier of the video containing the matching scene
timestamp	string	Timestamp where the scene occurs within the video
description	string	Generated description of the scene
score	float	Relevance score indicating how closely the scene matches the query

Semantic Search

Semantic search allows developers to search for meaning instead of exact keywords.

Example:

{
  "query": "scene where characters show strong emotional conflict"
}

The platform returns scenes whose narrative meaning matches the query.

This capability is powered by vector embeddings generated during intelligence analysis.

Metadata Filtering

Developers can combine semantic search with structured filters.

Example:

{
  "query": "emotional engagement scene",
  "filters": {
    "genre": "drama",
    "analysis_domain": "actor_emotion"
  }
}

This enables precise search across large video libraries.

Timestamp-Based Scene Retrieval

Search results include timestamps indicating where the identified scene occurs in the video.

Example timestamp:

00:02:34

Developers can use these timestamps to:

jump to specific scenes in video players
highlight key moments in applications
generate clips or previews

Hosted Search Experience

Vision AI Agents also provides a hosted search interface.

Developers may choose to use the hosted interface instead of building their own search UI.

In this model:

videos are ingested and analyzed
the hosted search UI retrieves indexed results
developers access results directly through the platform

This approach allows rapid deployment without building custom search infrastructure.

External Application Integration

Developers may integrate the search APIs directly into their own applications.

Typical integration flow:

Application
      ↓
Search API Request
      ↓
Vision AI Agents Search Engine
      ↓
Indexed Results Returned
      ↓
Application Displays Results

Search results can power features such as:

video discovery platforms
scene search tools
content recommendation systems
creative production tools

Search Best Practices

Developers building production search experiences should follow these best practices:

combine semantic queries with metadata filters
limit result sets for faster response times
cache frequent search queries
paginate large result sets

These practices improve search performance and user experience.