Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.lumiqtrace.com/llms.txt

Use this file to discover all available pages before exploring further.

The AI Hub brings together four AI-powered features that help you understand and improve your agent operations without manually sifting through raw data. Three live in the AI Hub’s tab interface — Cost Optimizer, Anomalies, and Ask. The fourth, Root Cause Analysis, is triggered directly from the trace flame graph.
Anomaly Detection and Natural Language Query require the Team plan or higher. Cost Optimizer and Root Cause Analysis require Pro or higher. The Free plan has access to the overview page only.
The Cost Optimizer analyzes your last 30 days of agent spend and produces a ranked list of specific, actionable recommendations for reducing your monthly bill. Each recommendation is generated by an AI model that examines your actual usage patterns — which models your agents use, how often, at what token volumes, and at what cache hit rates — so the suggestions are tailored to your project, not generic advice.Plan availability: Pro (weekly refresh), Team (daily refresh), Scale (real-time)

​
What each recommendation includes

  • Title — a concise description of the opportunity
  • Description — why this change would reduce cost based on your usage
  • Category — the type of optimization: model switching, caching, batching, prompt compression
  • Estimated monthly savings — a dollar figure for how much you could save
  • Confidence — how strongly your usage data supports this recommendation (recommendations below 60% confidence are filtered out)
  • Effort — low, medium, or high
  • Tradeoff — quality or performance considerations to review before implementing
  • Code example — an expandable snippet showing how to make the change
Recommendations are sorted by estimated savings so the highest-value changes appear first.
High-confidence, low-effort recommendations are your easiest wins — implement those first. High-effort recommendations often reveal architectural patterns in your agent system worth understanding even if you don’t act on them immediately.

​
Root cause analysis

Root cause analysis is triggered from inside a trace — not from the AI Hub. When you open a trace in the Traces page and click Explain with AI in the span detail panel, LumiqTrace analyzes every span in that agent run and returns a structured diagnosis. Plan availability: Pro (10/month), Team (50/month), Scale (unlimited)

​
What it returns

FieldDescription
SummaryPlain-English description of the full agent run
Root cause typeCategory: tool_failure, token_limit_exceeded, upstream_timeout, agent_loop, rate_limited, etc.
Problem spanThe specific span identified as the source of the issue
Contributing factorsOther spans or conditions that made the problem worse
Fix stepsPrioritized action list, each with an implementation note
Prevention tipOne recommendation for avoiding this class of problem

​
Timing and caching

Analysis runs asynchronously and takes 5–30 seconds depending on trace complexity — longer for multi-agent runs with many spans. Results are cached for 48 hours. Reopening the same trace shows the cached result instantly.
Root cause analysis uses an AI model with extended reasoning, which provides deeper diagnostic depth on complex agent failures. The extra latency reflects the model working through a multi-step reasoning process before producing its answer.