Cloudflare Adds Agent Tracing, with Truncation Limits and Uneven Payload Defaults
Cloudflare launched agent tracing, adding spans for agent invocations, model calls, tool runs, and approvals to existing Workers traces. Sessions replay turn by turn, though the docs warn traces are not lossless and payloads may be truncated. Payload recording defaults differ by framework, and from October 1, 2026 every span counts as a billable event. By Steef-Jan Wiggers
Cloudflare has introduced agent tracing for Workers, a feature that visualizes the interactions within deployed agent sessions. This feature includes agent-level spans, which were previously absent in Workers tracing. The new tracing functionality becomes payable starting October 1, 2026, under Workers Observability pricing.
Agent tracing captures agent invocations, model calls, tool execution, and approvals, with additional metadata such as model and token usage. Each turn generates a single trace, with subagent work nested under the parent agent's operation. The trace includes three fields: agent name, agent ID, and conversation ID. It is recommended not to derive the agent name from request or user identifiers to avoid multiplying distinct agents in the view.
There is a caveat regarding the approval span, which represents lifecycle events within a Worker invocation but does not measure human-wait latency across invocations. Additionally, teams should be aware that payload recording is not consistent by default, as certain data such as message or tool payloads are excluded unless explicitly enabled. The span attributes follow the OpenTelemetry Generative AI semantic conventions, and traces are exported to any OTLP endpoint.
Written by urgent.news from InfoQ's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
