OpenAI usage is easy to collect but hard to operationalize. Opsmeter adds endpoint and tenant ownership so teams can investigate spikes quickly.
OpenAI account totals do not tell operators which deploy, endpoint, or tenant changed cost behavior.
OpenAI account totals do not tell operators which deploy, endpoint, or tenant changed cost behavior.
Attribute OpenAI spend to endpoint, tenant, and user in one model.
Use warning/exceeded workflows tied to request-level context.
Catch prompt regressions with current-vs-baseline comparisons.
Read input/output token usage and preserve a stable externalRequestId across retries.
Include endpointTag, tenantId, userId, promptVersion, and environment on each ingest row.
POST request telemetry directly to Opsmeter ingest without introducing a gateway rewrite.
| Field | Example value | Why it matters |
|---|---|---|
| externalRequestId | req_6f12b5c | Dedupes retries and links traces to one cost event. |
| endpointTag | support.reply | Enables feature-level cost ownership and containment. |
| promptVersion | support_v9 | Detects deploy-linked cost/request regressions quickly. |
| tenantId | enterprise-alpha | Supports per-tenant profitability decisions. |
Send externalRequestId, endpointTag, promptVersion, userId, and tenantId with usage fields.
Keep app path unchanged while posting usage telemetry to Opsmeter ingest.
Use Top Endpoints, Top Users, and Prompt Versions to isolate change drivers.
No proxy rewrite required. Keep your current app path and enrich request metadata.