Vision AI Agents Developer Platform

APIs and architecture for ingesting, analyzing, indexing, and searching video intelligence.

Vision AI Agents enables developers and enterprises to ingest video content, run intelligence analysis, generate searchable indexes, and build video discovery applications powered by structured metadata and vector embeddings.

Platform Pipeline

The Vision AI Agents platform follows a structured intelligence pipeline.

Video Upload
      ↓
Video Ingest APIs
      ↓
Video Intelligence Analysis
      ↓
Optional Audience Testing
      ↓
Metadata + Vector Embeddings
      ↓
Search APIs
      ↓
Application Results

All downstream operations require the video_id generated during video ingest.

Quick Navigation

  • Platform Overview
  • Platform Workflows
  • Video Ingest APIs
  • Video Intelligence APIs
  • Audience Testing APIs
  • Search APIs
  • Getting Started
  • Platform Architecture
  • Authentication
  • Rate Limits & Usage Tiers
  • Error Handling
  • Search Integration Guide
  • Audience Testing Guide
  • Job Status & Polling Guide

Platform Overview

Vision AI Agents provides a complete pipeline for video intelligence:

  1. Video ingest
  2. Video analysis and signal extraction
  3. Metadata and vector indexing
  4. Search and retrieval
  5. Audience testing and engagement analytics

Developers can integrate the platform in two ways:

  • Use the Vision AI Agents hosted search and UI
  • Integrate Vision AI Agents APIs into their own applications

Platform Workflows

The platform supports four primary workflows.

1. Upload and Analyze

Developers upload one or many videos to the platform.

The system automatically:

  • generates a unique video_id
  • runs intelligence analysis
  • extracts structured signals
  • creates searchable indexes

Developers may request:

  • full analysis
  • selective analytics modules

2. Analyze Existing Videos

If videos already exist in the platform, developers can reference them using the generated video_id.

Example request flow:

  1. Send analysis request
  2. Provide video ID
  3. Select analytics modules

The platform returns structured analysis results.

3. Batch Video Ingest

Developers can ingest large video libraries using batch upload APIs.

The platform processes videos asynchronously and:

  • generates video IDs
  • runs analysis pipelines
  • queues processing jobs

Developers can check processing status using the job status polling endpoint.

4. Search Delivery

After videos are analyzed and indexed, developers can retrieve results using two models.

Hosted Search Experience

Developers use the Vision AI Agents hosted UI and search backend.

External Application Integration

Developers can integrate Vision AI Agents search APIs directly into their own applications.

Applications send search queries to the API and Vision AI Agents returns indexed results.

Video Ingest APIs

Video ingest APIs are the entry point to the platform.

All analysis and search operations require a system-generated video_id returned by the ingest endpoint.

Example endpoint:

POST /api/video/ingest

Capabilities include:

  • single video upload
  • batch video ingest
  • automatic video ID generation
  • automatic analysis pipeline initiation

Video Intelligence APIs

Video intelligence APIs allow developers to run analytics modules on previously ingested videos.

Example request:

POST /api/video/analyze

Developers can request:

  • full analysis
  • selective analytics modules

Video Intelligence Domains

Vision AI Agents extracts intelligence signals across multiple analysis domains.

Scene Actor Analytics

  • actor emotion engagement
  • actor eye contact engagement
  • actor attention intensity

Scene Elements Analysis

Scene element extraction includes:

  • audio genre recognition
  • audio stems and rhythm analysis
  • script linguistics analysis
  • color traversal patterns

Scene Psychology Analysis

Psychological signals extracted include:

  • color emotion classification
  • audio emotion classification
  • script sentiment classification
  • audience-to-actor emotional mirroring

Crescendo Detection

Vision AI Agents identifies narrative and emotional crescendos across video content.

Signals include:

  • color crescendo patterns
  • audio crescendo patterns
  • script crescendo patterns
  • scene crescendo synchronization
  • actor emotion synchronization

Audience Testing APIs

Audience testing must be explicitly requested through the API.

Audience testing requires a valid video_id generated by the ingest process.

Example request:

POST /api/audience/test

Request Parameters

Parameter Type Description
video_id string Video identifier returned from the ingest endpoint
participants integer Number of participants in the audience test (maximum 10)
analytics array List of requested audience analytics modules

Audience Analytics Modules

Module Description
emotion Measures aggregated emotional responses to video scenes
attention Measures viewer attention intensity and focus during scenes
engagement Identifies high engagement moments within the video
dropoff Detects scenes where viewer engagement declines

The platform returns aggregated engagement signals based on the selected analytics modules.

Search APIs

Search APIs allow developers to retrieve indexed video intelligence signals.

Example endpoint:

POST /api/search/query

Search capabilities include:

  • metadata search
  • vector similarity search
  • structured filtering

Search results can power:

  • Vision AI Agents hosted search UI
  • external developer applications

Authentication

All API requests require authentication using API keys.

Example header:

Authorization: Bearer API_KEY

Developers generate and manage API keys through the platform dashboard.

Rate Limits

API usage is governed by developer usage tiers.

Limits include:

  • concurrent requests
  • daily request limits
  • monthly token limits

Developers can upgrade tiers to increase throughput and processing capacity.

Getting Started

To begin integrating Vision AI Agents:

  1. Obtain API credentials
  2. Upload a video using the ingest API
  3. Run intelligence analysis
  4. Optionally request audience testing
  5. Retrieve indexed results through the search APIs