Skip to main content

Traces

Traces provide end-to-end distributed telemetry and performance tracing for agent execution flows and complex LLM tool calls. Powered by OpenTelemetry standards, Traces allow developers and engineers to debug multi-step agent reasoning, tool invocations, and model latency bottlenecks.

Navigate to OBSERVABILITY → Traces in the Aegis Console to inspect telemetry streams.

KPI Overview Cards​

Top KPI cards summarize telemetry health for the selected window:

  • Total Traces: Total trace count recorded.
  • Fail Rate: Percentage of traces ending in error status (green <5%, red >5%).
  • Avg Duration: Average trace execution duration in milliseconds (ms).
  • Total Tokens: Total tokens consumed across all spans in the traces.

Trace Filter Bar​

Use the multi-faceted filter bar to refine trace listings:

  • Agent: Filter traces by specific AI Agent name.
  • Team: Filter traces by team workspace.
  • Runtime: Filter traces by execution runtime environment.
  • User: Filter traces by user ID.
  • Status: Filter by execution status (OK or Error).

Click the Link icon in the header bar to copy a shareable URL to the current trace view with the exact time window pinned. Team members opening the link will see the identical time frame and filter configuration.

Trace List & Span Waterfall Detail​

Trace Table​

Each row lists high-level trace metrics:

  • Timestamp: Time when the root span started.
  • Trace ID: Unique OpenTelemetry trace identifier.
  • Agent: Agent handling the execution flow.
  • Spans: Total number of child spans (sub-steps) contained in the trace.
  • Duration: Total execution duration in milliseconds (ms).
  • Tokens: Total tokens consumed.
  • Status: Overall execution status (OK or Error).

Span Waterfall Detail Page​

Clicking any trace opens the dedicated Trace Detail Page (/observability/traces/:trace_id):

  • Span Timeline Waterfall: Visual interactive Gantt chart displaying parent and child spans (LLM generation, vector retrieval, python tool execution).
  • Span Inspector Panel: Click individual spans to view attributes, execution events, raw input/output JSON payloads, model parameters, and exception tracebacks.