LLM Cost

LLM Cost lives under Settings → Admin → LLM Cost. Use this page when the model bill jumps: it shows which service, which model, and which feature is spending the tokens. Both the dashboard (finding narratives, verdicts) and the Query assistant write the same usage rows, so this is the whole product’s spend. Requires the view audit permission.

Time range

A segmented toggle for 7 days, 30 days, or 90 days. The default is 30 days.

Summary

Four hero metrics at the top:

Metric	Meaning
Spend	Priced USD cost in the selected range, plus how many calls were priced
Total tokens	Prompt (in) plus completion (out)
Output share	Percent of tokens that produced an answer; the rest is input. Also shows what share of input was served from cache
Failed calls	Errors in range. Retries still cost tokens

If some calls ran on a model with no price-book rate, those tokens are counted but excluded from the dollar figure, and a banner calls this out. Calls that returned no usage block from the provider are excluded from both tokens and cost.

Daily spend by service

A stacked bar chart of priced spend per day, split by the Hilt service that called the model (for example the dashboard backend versus the Query service). Unpriced models are excluded from these bars.

Breakdowns

Four tables, each with share, call count, tokens, and cost:

Table	What it answers
By page	Which screen triggered the spend. Background workers have no page. Also shows average prompt tokens per call
By service	Which Hilt service is calling the model
By model	Which model the spend landed on (the gateway’s actual routed id), plus provider
By feature	Which product feature spent the tokens

Settings map → Settings
Natural-language assistant → Query
Subscription invoice → Billing

Findings

Estate

Settings

Time range

Summary

Daily spend by service

Breakdowns

Time range

Summary

Daily spend by service

Breakdowns

Related