AI Center / GenAI Semantic Conventions

Coralogix AI Center is a trace product surface. It does not populate from logs or metrics alone, and customer-visible rows depend on GenAI span attributes being present in the trace data path that AI Center reads.

Use this reference when a customer has LLM traces in Coralogix but AI Center rows, model/provider columns, prompts, or token usage are missing.

Detection attributes

AI Center supports both legacy and newer OpenTelemetry GenAI semantic convention shapes.

If none of those attributes are present on the GenAI span, AI Center should be treated as having no GenAI row to render even if the trace itself exists.

Product path requirements

• AI Center consumes trace data. Logs and metrics do not create AI Center LLM rows.
• AI Center depends on the trace/archive path it is designed to read. A trace that exists only in a frequent-search or non-archive path can be visible elsewhere while absent from AI Center.
• Gateway-side AI enrichment can route only the GenAI spans from a mixed trace for enrichment, then return the enriched OTLP trace data to the normal trace pipeline. The detection key still comes from GenAI span attributes.
• Preserve trace identity and Coralogix ingestion headers through any enrichment side path. Header/topic wiring belongs to the ingestion or collector owner, not this skill.

Required troubleshooting checks

1. Confirm the spans are traces, not logs or metrics.
2. Inspect a raw span and look for gen_ai.system, gen_ai.provider.name, or gen_ai.input.messages.
3. Check whether message content is disabled by instrumentation defaults. Common toggles include:
   • OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true
   • TRACELOOP_TRACE_CONTENT=true
4. For newer GenAI semconv, check whether the process opted in with OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental.
5. If the customer uses a proxy such as LiteLLM, confirm the proxy or custom instrumentation emits GenAI span attributes. A proxy can forward LLM traffic without automatically producing AI Center-compatible spans.

Common attribute-shape mistakes

Ownership boundary

This skill owns the telemetry-shape diagnosis: which gen_ai.* attributes are required, which signal is consumed, and why a proxy/custom library might miss them.

Hand off:

• SDK/library setup and manual span code to opentelemetry-instrumentation.
• Collector routing, archive/exporter placement, and gateway side topics to opentelemetry-collector or the owning ingestion workflow.
• OTTL transforms to opentelemetry-ottl.

References

• OpenTelemetry GenAI semantic conventions: https://opentelemetry.io/docs/specs/semconv/gen-ai/
• apm-service-catalog.md for general span/resource identity.
• associated-boundaries.md for topology handoffs.