Skip to main content

​
Overview

Trio exposes metrics in Prometheus format at /metrics. You can scrape this endpoint with a Prometheus server and visualize the data in Grafana.

​
Key Metrics Explained

​
Provider Calls

external_provider_calls_total Shows how many times each VLM provider was called. Use this to track costs and detect API key issues.

​
Job Statistics

  • jobs_created_total
  • jobs_completed_total
  • jobs_failed_total
Use these to monitor job success rates and reliability.

​
Frame Statistics

frames_processed_total Tracks captured vs skipped frames. High skip rates indicate the pre-filter is working effectively on static scenes, saving costs.

​
Webhook Statistics

webhooks_sent_total Monitor status="failed" to detect if your webhook receiver is down.

​
Cost Analysis

To estimate costs, track external_provider_calls_total for each provider and multiply by the provider’s cost per call. Example:
  • Gemini 2.5 Flash: ~$0.00002 / call
  • GPT-4o Mini: ~$0.000015 / call

​
Optimization Tips

  1. Use Pre-filtering: Keep enable_prefilter set to true (default) to skip static frames.
  2. Adjust Interval: Use a 30-60s interval for most monitoring use cases.
  3. Use Check-Once: For single queries, use /check-once instead of starting a job.