Skip to main content

​
Overview

The Live Digest endpoint captures frames from your stream and generates narrative summaries. The quality and cost of summaries depend on how you configure frame capture: how often frames are captured and over what time window.

​
Key Parameters

​
window_minutes

How long to accumulate frames before generating a summary.
  • Shorter windows (5-10 min): Frequent, detailed summaries. Better for fast-moving events.
  • Longer windows (30-60 min): Less frequent summaries. Good for gradual trends.

​
capture_interval_seconds

How often to capture a frame from the stream.
  • Frequent capture (30 sec): More frames, richer detail. Higher costs.
  • Sparse capture (180+ sec): Fewer frames, lower cost. Works well for slow-changing scenes.

​
Configuration Examples

​
Traffic Monitoring

Check for incidents throughout the day: 
{
  "window_minutes": 10,
  "capture_interval_seconds": 60
}
Result: A new summary every 10 minutes with captured frames

​
Weather Tracking

Hourly weather summaries: 
{
  "window_minutes": 60,
  "capture_interval_seconds": 300
}
Result: Summaries every hour, showing overall weather trends

​
Security Monitoring

Detailed 5-minute snapshots: 
{
  "window_minutes": 5,
  "capture_interval_seconds": 30
}
Result: Frequent detailed summaries for security events

​
Low-Cost Monitoring

Daily trend reports with minimal API calls: 
{
  "window_minutes": 1440,
  "capture_interval_seconds": 3600
}
Result: One summary per day covering 24 hours

​
Cost Optimization

Cost = Number of frames sent to VLM × Price per VLM call Lower costs by:
  • Increasing capture_interval_seconds - Capture fewer frames (e.g., 180 instead of 60)
  • Increasing window_minutes - Summarize longer periods between summaries
  • Using faster models - gpt-4o-mini or gemini-2.5-flash cost less than claude-3-5-sonnet
Trio automatically optimizes processing within your configuration to balance cost and summary quality.

​
Output

Each summary includes:
  • summary: Narrative description of what happened during the window
  • frame_count: Number of frames analyzed
  • window_minutes: The time period covered
  • grid_b64: Base64-encoded composite image (if include_frames: true) showing the frames analyzed

​
Getting Started

  1. Start with the use case that matches yours above
  2. Check the summary quality and costs in your metrics
  3. Adjust window_minutes and capture_interval_seconds based on your needs
  4. Use shorter intervals for detail, longer intervals for cost savings