Usage Tracking

Every API request is tracked with detailed usage metrics including tokens, cost, and latency.

Per-Request Tracking

Every response includes a usage object:

{
  "usage": {
    "tokens_prompt": 25,
    "tokens_completion": 150,
    "tokens_total": 175,
    "cost": 0.0052,
    "latency": 1200.5
  }
}

What's Tracked

Metric	Description
tokens_prompt	Input tokens sent to the model
tokens_completion	Output tokens generated by the model
tokens_total	Sum of prompt + completion tokens
cost	Cost in USD for this request
latency	Processing time in milliseconds
cache_read_tokens	Tokens served from response cache
cache_write_tokens	Tokens written to response cache
reasoning_tokens	Internal reasoning tokens (supported models)
web_search_count	Web searches triggered

Credit System

• Each account has a credit balance in USD
• Credits are deducted based on the cost of each request
• Minimum credit requirements vary by endpoint type
• BYOK requests do not consume credits
• Check your balance through the IndoxHub dashboard

In-stream usage events

Streaming responses emit the same totals inside the stream as a named usage_final SSE event, so clients don't need a second API call to retrieve billing data after a stream ends:

event: usage_final
data: {"type":"usage_final","input_tokens":15,"output_tokens":1,"cost_usd":2.85e-06,"latency_ms":4693}

The fields map 1:1 to the per-request usage object documented above (input_tokens ≡ tokens_prompt, output_tokens ≡ tokens_completion, cost_usd ≡ cost, latency_ms ≡ latency). See SSE Events for the full per-event reference.

Storage

Usage data is stored in two layers:

• MongoDB: Detailed per-request logs with full request/response data
• PostgreSQL: Aggregated daily summaries for efficient reporting