Historical AI Analysis

Overview

Historical AI Analysis is a conversational interface that lets you ask natural language questions about your fleet's historical data. Powered by Claude, it translates your questions into database queries, analyzes the results, and provides narrative insights with optional inline charts.

Access it from the Reports page by clicking the AI Analysis tab. Requires a paid subscription (Professional tier or higher). Each analysis uses 1 AI query from your subscription pool.

How It Works

Ask a question — Type a natural language question like "Which agents had the highest CPU usage last week?" or "What are the most common error events in the past 24 hours?"
Query planning — Claude examines your question and your organization's data schema to determine which database queries are needed. It may run multiple queries to gather the information required for a complete answer.
Streaming response — Results stream back in real-time via Server-Sent Events (SSE). You'll see the narrative text appear progressively as it's generated.
Charts and follow-ups — If the data warrants visualization, an inline chart is automatically generated. Follow-up question suggestions appear at the bottom of each analysis to help you explore further.

Example Questions

"Which agents had the highest CPU usage last week?"
"What are the most common error events in the past 24 hours?"
"How does this week's fleet health compare to last week?"
"Show memory usage trends across all agents for the past month"
"Are there agents that consistently have both high memory and high disk I/O?"
"Which ETW providers generate the most events?"
"How many diagnostic sessions were run last month?"
"Which agents are most likely to need hardware upgrades based on current trends?"

The system has access to the same data sources as the Data Explorer: health metrics, events, event batches, correlation analyses, detected patterns, and live sessions.

Follow-Up Chains

Each analysis can lead to follow-up questions. When you ask a follow-up, the previous analysis context (your original question and a summary of the findings) is sent along with the new question. This enables deeper, multi-step exploration like:

"Show me CPU trends for the past week" →
"Which specific agents are driving the spikes?" →
"What events were happening on AGENT-05 during the spike at 2 PM Tuesday?"

Analysis History

Switch to the History tab to view past analyses. Each entry shows the question asked, a summary of the findings, when it was run, and key metrics (data points analyzed, execution time). Click any history entry to reload the full analysis with its chart and follow-up suggestions. History is paginated with 10 entries per page.