OverviewReal-time model health across all deployments

Latency percentiles

P50 / P90 / P99 + TTFT over time

Throughput

Requests / min
req/minprev period

Token usage

Input vs output tokens over time
InputOutput

Quality signals

Live eval scores · higher is better

Sample trace

Request lifecycle · embed → retrieve → rerank → LLM → tool → response
trace_—

Latency percentiles

End-to-end response time distribution over the selected window

Latency distribution

Response time histogram (ms)

Time to first token

TTFT by model

Spend trend

Daily cost across all models

Cost by model

Spend this period · USD

Token usage

Input vs output tokens stacked over time
Input tokensOutput tokens

Eval gauges

Rolling quality metrics · selected model

Model comparison

Head-to-head eval scores

Hallucination & toxicity trend

Flagged-response rate over time · lower is better
HallucinationToxicityRelevance drop

Trace waterfall

Span timeline for a sampled request · click a log row for others

Request logs

Time Request ID Model Tokens Latency Cost Eval Status
Neuron · demo build · synthetic telemetry · designed by Gideon Awolesi
Designed by Gideon Awolesi