SentinelGateway Request Tracing — FinOps Observability

SentinelGateway logs every LLM request — successes and failures alike — with a stable trace ID, model, latency, token counts, dollar cost, and governance flags (cache hit, fallback, PII redaction). The Command Center provides live-tail trace streaming, status and model filters, prompt substring search, cursor pagination, and per-model cost breakdowns with today and month-to-date spend aggregation.

📊 GOVERNANCE & FINOPS — End-to-End Observability

Every token. Every cent. Every trace.

Stable trace IDs, per-model cost attribution, and live telemetry across your entire AI fleet — including the failed requests other tools quietly drop.

No credit card required · Drop-in OpenAI SDK compatible · Live in 60 seconds.

Command Center · live traces

LIVE
tr_9f2c41a7
200 SEMANTIC SCRB
24ms
tr_8b1e90d2
200 FALL
812ms
tr_77aa03fc
ERR SCRB
429 · 0ms
today's spend $0.00

How it works

Observability is captured in-line on the request path — not sampled, not best-effort, not dropped on error.

1

Ingest & Normalize

Every request is stamped with tenant, model, and timing metadata at the edge. Quota checks, rate limits, and auth outcomes are captured before routing begins.

2

In-Flight Evaluation

Cache hits, semantic matches, PII redactions, and fallback events are flagged on the trace as they happen — stacked as governance badges, never overwriting each other.

3

Delivery & Audit Logging

Tokens, cost, status code, and latency persist to the audit log with a stable trace ID — for 200s, 429s, and 5xxs alike. Filter, search, and paginate in the Command Center.

Governance your CFO can read

FinOps-grade attribution without a data pipeline, a warehouse, or a BI team.

100%

Request coverage

Errors and rate-limits logged with zeroed cost — nothing dropped.

3s

Live-tail cadence

Traces stream into the dashboard as requests complete.

$0.01

Cost granularity

Cent-level attribution per model, per request, per tenant.

30d

Spend windows

Today and month-to-date rollups beside every trace stream.

See your first trace in under a minute.

Sign up in 60 seconds. Every request is traced from the very first call — no agents to install.

Explore the platform