Use Cribl Search to explore, investigate, and visualize LLM telemetry.

When LLM telemetry lands in Cribl Search, you can:

• Run ad hoc investigations (errors, latency, usage spikes).
• Correlate LLM spans with infrastructure or security telemetry.
• Build dashboards and scheduled searches for usage, cost, and performance.

Prerequisites ​

• LLM telemetry in Cribl Search from your instrumented application.
• Datasets configured in Cribl Search for LLM data.

Search LLM Telemetry ​

Store your LLM telemetry in a Dataset, then query it directly in Cribl Search with KQL for fast investigation and analysis.

• Create a Dataset and use Dataset rules to route your LLM telemetry into it. Search Datasets are hosted in lakehouse engines and are optimized for fast, schema-aware search and AI workflows.
• Use that Dataset in queries to filter by service, model, or environment; slice by token usage or cost; and narrow on error or status fields for failing requests.

Investigate Incidents and Performance Regressions ​

During an incident (for example, elevated LLM error rates or latency):

• Start from an LLM telemetry Dataset in Search.
• Use filters and aggregation operators (such as summarize or timestats) to break down failures by model or deployment, prompt type or application feature, and tenant, user, or region.
• Correlate with other Datasets (for example, infrastructure or API gateway logs) using join or related operators to see upstream and downstream symptoms together.

Build Dashboards for LLM Usage and Cost ​

Use Cribl Search Dashboards to visualize:

• Requests over time by model or application.
• Token usage and derived cost over time.
• Error rates by environment or tenant.

Add charts for P95/P99 latency, token count distributions, and traffic share per model or feature path.

Automate Checks with Scheduled Searches and Notifications ​

Turn critical queries into Scheduled Searches–for example, alert when estimated LLM cost exceeds a threshold over the last hour, or when error rate for a model exceeds a limit over 10 minutes. Attach Notifications for email, Slack, SNS, or other targets.

Related Topics ​

• Route LLM telemetry to multiple destinations
• Emit LLM cost and usage metrics from token counts