AI Gateway Observability
AI Gateway logs every LLM request and response, giving you full visibility into model usage, costs, latency, and guardrail results across all workspaces.
| Where Can I Use This? | What Do I Need? |
|---|
- Prisma AIRS AI Gateway (Americas region)
|
- AI Gateway activated
- At least one LLM integration and workspace configured
|
Without observability, AI usage is a black box — you can't tell which teams are spending
the most on LLM calls, which models are responding slowly, or whether guardrails are firing.
AI Gateway solves this by recording every transaction as a log entry that includes the
workspace, model, provider, token count, cost in flex credits, request latency, cache status,
and the full request and response payloads.
Logs are retained for one year and are accessible in Strata Cloud Manager under AI Security
→ AI Gateway → Observability → Logs. Each log entry has two views:
- Metadata panel (left) — sourced from Strata Cloud Manager; shows workspace, model,
provider, gateway API key used, token counts, cost, latency, and Config Status (whether a
routing config was applied and whether the response was a cache hit).
- Request/response panel (right) — sourced from the gateway data plane; shows the
full request prompt and model response in both Original (as sent) and Transformed (after
guardrail redaction) forms.
AI Gateway charges 1 flex credit per 4 characters of LLM input or output. The cost shown
in logs reflects this conversion. Budget consumption toward workspace limits is tracked
per-request and is visible in the log metadata. Logs are exportable via OpenTelemetry for
integration with external SIEM or observability platforms. For Hybrid deployments, logs can
also be exported to Amazon S3.
