Dashboard

By feature

The feature breakdown groups every LLM call by its feature tag and shows you exactly which parts of your app are driving your bill.

What you'll see

• Total cost — cumulative spend for this feature
• % of bill — what share of your total LLM spend this feature represents. Highlighted in red when over 50%.
• Calls — total number of tracked calls for this feature
• Avg tokens — average total tokens per call. High numbers here usually mean large context windows or long prompts.
• Avg cost / call — average spend per individual call. Compare this across features to find efficiency gaps.
• Avg latency — average response time in milliseconds

What to look for

A large gap in avg cost / call between features is a signal worth investigating. If search costs 10x more per call than chat, the likely cause is context window size — how many tokens you're passing in with each request.

Setting feature tags

Feature tags come from the feature parameter in your SDK call:

await trackedCall({feature: 'search',  // this is the tag...})

Use consistent, descriptive names. The feature breakdown is only as useful as the tags you define.

Next: By user →