Skip to main content
Version: 6.1

Data Sources

The AI Observability module collects AI environment telemetry from multiple sources and normalizes it into a unified set of gen_ai_* runtime indexes based on the OpenTelemetry GenAI data model. Normalized data is used by dashboards, inventory, and global Service Monitor Toolkit metrics.


Connected Sources

SourceTypeCollection RouteWhat is collectedWhere it goes
DCGM Exporter /metricsGPUSAF Beat Scriptbeat - LogstashGPU util, memory, temperature, ECC, XID errors, energy consumptiongen_ai_gpu_metrics
vLLM/metricsInference runtimeSAF Beat Scriptbeat - LogstashQueue depth, latency, KV cache, prefix cachegen_ai_metrics
Claude Code (.claude/projects)Local clientSAF Beat Filebeat - LogstashRequests, responses, tokens, session eventsgen_ai_events
Claude Desktop (Library/Logs/Claude)Local clientSAF Beat Filebeat - LogstashClaude Desktop application events and errors (macOS)gen_ai_events
Codex (.codex/sessions)Local clientSAF Beat Filebeat - LogstashRequests, responses, tokens, TTFT, reasoning, cachegen_ai_events, gen_ai_cost
LiteLLM_SpendLogs PostgreSQLLLM gatewayLogstash JDBCTokens, cost, provider, model, durationgen_ai_cost, gen_ai_events
AI agents OTLP tracesAI agentOTLP-domainAgent steps, tool calls, spans, errorsgen_ai_traces
AI agents OTLP logsAI agentOTLP-domainQuestions, answers, diagnosticsgen_ai_events
AI agents OTLP metricsAI agentOTLP-domainRequest rate, error rate, custom latencygen_ai_metrics
LiteLLM OTLP callbackLLM gatewayOTLP-domainRequest traces, latency, routinggen_ai_traces

Index Model

The module uses a unified runtime index set with wildcard patterns:

IndexTime fieldSourcesPurpose
gen_ai_events*@timestampLiteLLM, AI agents, Claude Code, CodexLLM requests, agent events and local client events, responses, errors
gen_ai_traces*startTimeLiteLLM OTel, AI agentsTraces and spans of AI requests and agent steps
gen_ai_metrics*@timestampvLLM, AI agentsPerformance metrics of inference services and agents
gen_ai_cost*@timestampLiteLLM SpendLogs, CodexCosts and usage by models, providers, commands
gen_ai_gpu_metrics*@timestampDCGM ExporterHardware GPU metrics: load, memory, temperature, errors

Notes

  • ISM policies manage index lifecycle (rollover, retention) and are included in the content module
  • SA Data Collector, SAF Beat, and otlp_collector configurations are included in the data collection configurations