AI Insights

AI-generated recommendations to optimize your usage

AI Insights analyzes your last 30 days of usage and generates personalized recommendations. Rather than reading raw charts, you get a structured report written by an AI that has reviewed your actual data — costs, model choices, token patterns, peak hours, and latency — and identified concrete opportunities to improve.

How it works

Insights runs on demand. Open the Insights page and click Run Analysis. The analysis reviews the following data from the past 30 days:

• Total tokens consumed and total cost
• Breakdown by model (tokens, cost, input/output ratio)
• Average tokens per request
• Usage trend vs. the preceding 30 days
• Peak usage hours
• Latency per model (average, P95, max)

The result is cached and displayed immediately on your next visit — you don’t need to re-run the analysis to see it again.

Insight categories

Each analysis produces a report organized into four categories:

Cost optimization

Identifies where you’re spending the most and where you could spend less. This includes spotting high-cost models that could be replaced by cheaper alternatives for certain task types, flagging inefficient input/output ratios, and highlighting requests where the token count is higher than expected.

Model selection

Reviews which models you’re using and whether they’re the right fit for your workloads. You might see recommendations to route simpler tasks to lighter models, or to consolidate usage across fewer models to reduce overhead.

Usage patterns & behavior

Surfaces patterns in how and when you use AI — peak hours, requester distribution, bursts of activity, and any unusual behavior worth investigating. Useful for capacity planning and understanding how your applications actually consume AI.

Projections & trends

Extrapolates your current trajectory to give you a sense of where costs and usage are heading. Includes month-on-month comparisons and projected monthly spend based on recent activity.

Each recommendation is tagged with a severity level:

Level	Meaning
Positive	Something working well
Info	Neutral observation
Warning	Something worth your attention
Critical	Significant issue or cost risk

Plan limits

Plan	Analyses allowed
Free	1 per month
Pro	5 per day

The Free plan counter resets on the first of each month. If you’ve used your one monthly analysis, the page shows a banner with the option to upgrade to Pro.

Run your first analysis after you’ve accumulated at least a week of data through the proxy. With less than a few days of history, the recommendations will be limited — the more requests and models the analysis has to work with, the more specific and actionable the output.