Viewing Traces

The traces page

The Traces page (/traces) shows all traces in the current project, ordered by most recent first. Each row shows:

• Name — The trace name (e.g., answer-question, summarize-document)
• Span count — Number of spans in the trace
• Duration — Total time from first span start to last span end
• Cost — Sum of estimated costs across all spans
• Tokens — Total input + output tokens
• Status — Green if all spans completed, red if any span failed, yellow if any span is still running
• Time — When the trace was created (relative, like "2 minutes ago")

Click a trace row to navigate to its detail view.

Trace detail

The trace detail page (/traces/:id) shows the span tree for a single trace. The layout has two panels:

Left panel: span tree

A collapsible tree view of all spans in the trace, showing the parent-child hierarchy. Each span shows:

• Name — The span name
• Kind — Icon or label indicating the kind (llm_call, custom, etc.)
• Duration — A horizontal bar showing relative duration
• Status — Color-coded: green (completed), red (failed), yellow (running)
• Cost — Estimated cost (for llm_call spans)

Click a span to select it and show its details in the right panel.

Right panel: span details

When a span is selected, the right panel shows:

• Overview — Name, kind, status, duration, timestamps
• Input — The full input data as formatted JSON. For LLM calls, this is typically the message array.
• Output — The full output data as formatted JSON. For LLM calls, this is the response text or structured output.
• Kind details — For llm_call spans: model, provider, input tokens, output tokens, cost. For custom spans: kind string and attributes.
• Error — If the span failed, the error message is shown prominently.
• Actions — "Export to Dataset" button to save this span's data as a datapoint.

Timeline view

The span tree includes a timeline visualization showing when each span started and ended relative to the trace. This makes it easy to see:

• Which spans ran in parallel vs. sequentially
• Where the most time was spent
• Gaps between spans (time spent in your application code, not in traced steps)

Deleting traces

From the traces list, you can delete individual traces or clear all traces. Deleting a trace also deletes all its spans.

From the API:

# Delete a single trace
curl -X DELETE "https://api.traceway.ai/api/traces/${TRACE_ID}" \
  -H "Authorization: Bearer tw_sk_..."

# Delete all traces
curl -X DELETE "https://api.traceway.ai/api/traces" \
  -H "Authorization: Bearer tw_sk_..."

Deletion is immediate and irreversible.

Data export

Export all data (traces, spans, datasets, datapoints) as a JSON file:

curl "https://api.traceway.ai/api/export/json" \
  -H "Authorization: Bearer tw_sk_..." \
  -o traceway-export.json

This is useful for backups or migrating data between Traceway instances.