The agent trace, end to end.
One root span per agent run. Every LLM call, tool call, sub-agent spawn, retry, and outcome is a child span. Schema-validated, OpenTelemetry-canonical, replayable.
Your agent stops being a black box.
Most teams ship agents and find out what happened by reading the response. Trefur turns the agent into a live map — every node is a decision, a tool call, or a retry, and the trace pulse shows you exactly which path the run took, in what order, and at what cost. This is a real agent flow, simulated below.
How it works.
Drop in the Trefur SDK, or run the trefur collector on your network, or simply add Trefur as a second OpenTelemetry exporter alongside whatever you already run. The data path is unchanged; Trefur reads the GenAI semantic conventions that your runtime already emits.
Every agent run becomes a single trace. The framework you use decides the shape of the tree. LangChain emits one root and chain steps as children. CrewAI emits crew, task, and hand-off spans. Pydantic AI emits typed function-call spans. MCP emits server, client, tool-call, and resource-fetch spans. Custom orchestrators that emit OTel GenAI spans show up the same way.
Traces are searchable by trace ID, by agent name, by customer interaction, by error class, by time window. The trace view shows the tree on the left and span details on the right — inputs, outputs, tokens, latency, cost, and any errors.
What you can do with it.
A trace, in your terminal.
Works with.
Get your first agent trace.
Free tier. No card required. First trace in under five minutes.