Observability
Every request logs its provider, model, tokens, cost, latency, fallback, and cache status — surfaced live in the dashboard and on response headers.
You cannot govern what you cannot see. Every request through the gateway is logged with the full routing decision and economics, so you can trace any call and watch cost and latency in real time.
Per-request transparency headers
Each response carries the routing decision inline:
| Header | Meaning |
|---|---|
X-Nyuro-Model | The model that actually answered |
X-Nyuro-Route-Reason | Why that model was chosen |
X-Nyuro-Fallback-Used | true if the primary candidate was skipped |
X-Nyuro-Candidates | The full candidate chain that was resolved |
What every request logs
For each call the gateway records the provider, model, prompt/completion tokens, cost in USD, latency, whether a fallback fired, and whether the response was served from cache. Cache-served requests are flagged so they do not distort provider latency or cost analysis.
Dashboard metrics
The console aggregates these logs into live metrics:
- Volume — total requests and requests in the last 24h
- Reliability — success rate and fallback rate
- Performance — average latency (and per-model average latency, used to sort the model catalog)
- Cost — total spend, broken down by provider and model
Metrics are computed from the same request log that powers budgets, so what you are billed for and what you see in the dashboard always agree.
Programmatic access
The same data is available from the API for your own dashboards and alerting:
# Aggregated dashboard metrics
curl https://api.nyuro.ai/api/v1/metrics/dashboard \
-H "Authorization: Bearer neu_live_…"
# Recent request logs
curl "https://api.nyuro.ai/api/v1/logs?limit=100" \
-H "Authorization: Bearer neu_live_…"