Trace explorer
A calm LLM observability console: p50 and p95 latency over 24 hours, a brush that filters the runs to a spike, and a waterfall that zooms into a span while its time axis rescales, with the span's detail and a one line AI read of where the time went.
Data and insightNo. 2392 variationsFree with an account
Playing on its own
Variation
- The idea
- Production agents are judged on latency, cost and quality together; an AI that reads the trace for you, and a waterfall you can zoom into, turn a spike into a cause
- Autoplay
- The cursor brushes the 14:00 spike and the runs filter to four slow ones; it opens the slowest and the waterfall draws as the summary streams (71 percent in retrieval); it clicks the 2.4 s vector search retry and the waterfall zooms so that span fills the width, the axis rescales and the span detail slides in; it zooms back out from the breadcrumb, shows cost per span, and clears the filter.
- When you click
- Drag across the chart to filter (handles are sliders: arrows, Shift for 2 h, Delete clears); the spike button jumps to the peak; sort runs by time or duration and open one with Enter; click or Enter a span to zoom, the breadcrumb or Esc to zoom out, Left goes to the parent; the Cost switch changes the labels. Phone: chart on top, runs as a list, the waterfall in a bottom sheet.
- Look
- flat hairline cards, neutral 8 px bars, brush, zoom with scale transforms, sans with mono timings type
- Variations
- Aurora, Daylight
- Files
- index.html, style.css, script.js and prompt.md. No framework, no CDN, works offline.
The prompt and the files
Paste the prompt into Claude, ChatGPT, Lovable or any coding agent to rebuild it in your own product, or take the three files as they are.
Loading the files