Skip to main content
The Traces page (/ide/observability/traces) is where you browse everything Oxygen has recorded — every agent run, workflow execution, and analytics run. Beyond scrolling the list, it gives you a set of controls to find the exact run you care about, watch runs land in real time, and compare two runs side by side.
The Traces page is enterprise-only. Start the server with --enterprise and set a storage backend (OXY_OBSERVABILITY_BACKEND) — see Observability for the full setup.

​
Searching

Type in the search box to filter traces by:
  • Span name — e.g. agent.run_agent, analytics.run
  • Agent reference — the agent that produced the run
  • Prompt / question — the text the run was asked to handle
  • Trace ID — paste a full trace id to jump straight to it
Search is case-insensitive and matches substrings, so revenue finds any run whose prompt mentions revenue. It updates as you type (after a brief pause) and combines with every other filter below.

​
Filtering by status

The status dropdown narrows the list to:
  • All statuses — everything (default)
  • Success — runs that completed cleanly
  • Errors — only runs that ended in an error
Pick Errors to triage failures without the noise of successful runs.

​
Choosing a time range

By default the list shows the last 30 days. Use the range control at the top right to change the window:
  • Presets — 1h, 24h, 7d, 30d, 90d rolling windows.
  • Custom — click Custom and pick an absolute From and To date and time. This is the one to use when you’re chasing an incident that happened at a known time. An absolute range overrides the preset until you pick a preset again.
Traces are retained for 90 days, so 90d is the widest window available.

​
Live-tail

Toggle Live on to auto-refresh the list every few seconds. New traces appear at the top as runs complete — useful while you’re actively reproducing an issue or watching a deploy. A small pulsing dot indicates live mode is on. Toggle it off to freeze the view.

​
Card vs. table view

Use the view toggle to switch between:
  • Cards — a roomy view with the prompt, status, type, agent, duration, and token count for each run. Best for reading.
  • Table — a dense, scannable grid of the same runs. Best for eyeballing many traces at once, sorting through a spike, or comparing durations down a column.
Your choice applies to the current filters and either view links straight into the Trace Detail page when you click a row.

​
Comparing two traces

To understand why one run was slower — or cost more — than another, compare them:
  1. Check the box on two traces (in either card or table view).
  2. Click Compare in the selection bar that appears.
  3. A dialog shows both runs side by side — status, total duration, span count, LLM/tool calls, and token totals — with a B − A difference row that highlights exactly how the second run diverged from the first.
From the compare dialog you can click Open on either side to jump into its full trace detail.
Compare a fast run against a slow one for the same question to spot where the extra time or tokens went — often an extra LLM round-trip or a heavier SQL step.