Real-time cost analytics with time-series charts, per-model breakdowns, performance metrics, and configurable date presets.
Overview
The Cost Dashboard gives you full visibility into your LLM spending and performance. It aggregates data from all enrichment records in your organization and presents it through interactive charts and summary cards. Use it to identify cost trends, compare model efficiency, and optimize your enrichment pipeline.
Tabs
2
Date Presets
4
Chart Types
5
Cross-Org
Admin
Date Presets
Select a time range from the sidebar. The URL updates to reflect the selected preset (e.g., /costs/30d), enabling bookmarkable views:
Preset
Time Range
Chart Grouping
7d
Last 7 days
Daily
30d
Last 30 days
Daily
90d
Last 90 days
Weekly
all
All time
Monthly
System administrators get an organization selector in the sidebar with three settings: their own organization, all organizations aggregated, or one named organization. Everyone else — owners included — always sees their own organization's costs; the backend resolves the scope, so the filter cannot be worked around from the client. The choice is persisted per user in local storage.
Cost Overview Tab
The default tab provides a comprehensive spending breakdown:
Summary Cards
1Everything spent in the period
2Mean cost of one call
3Mean cost of one filled property
The two averages are what make runs comparable: cost per request answers “what does one enrichment cost”, cost per property answers “what does one filled field cost” — the number that moves when a schema grows.
Charts & Tables
Cost Over Time— Line chart showing spending trends across the selected period, grouped by day/week/month.
Cost by Model— Horizontal bar chart of the top 10 models by total cost. Quickly identify the most expensive models.
Token Usage— Breakdown of input tokens, output tokens, and total tokens consumed.
Model Insights— Cards highlighting the most used model and the most expensive model.
Daily Breakdown— Table with date, request count, total cost, average cost per request, and token totals.
One period, four readings of the same spend: the total, its shape over time, which model carries it, and the day-by-day detail.
Performance Analysis Tab
Switch to the Performance tab to analyze model efficiency and identify cost-performance trade-offs:
Summary Cards
Total Records
156
Enrichment records in the period
Models Used
8
Distinct models that produced results
Language Variants
3
Languages used in multilingual enrichment
Token Ranges
4
Distinct input token size buckets
Charts & Tables
Cost vs Duration— Scatter chart with bubble sizes proportional to request count. Each bubble is a model — find the best balance of speed and cost.
Performance by Model— Table comparing request count, average cost, average duration, and token statistics per model.
Cost by Language Count— Bar chart showing how cost scales with the number of languages selected for multilingual enrichment.
Cost by Input Token Range— Bar chart breaking down costs by input prompt size buckets (e.g., 0–1K, 1K–5K, 5K–10K tokens).
Performance by Schema Property Count— Table showing how enrichment cost and duration correlate with schema complexity (enrichment records only).
The chart is a choice, not a report: the model you want is usually the one closest to the bottom-left corner that still meets your quality bar.
Optimization Tips
Compare Models
Use the Cost vs Duration scatter chart to find models that deliver good quality at lower cost. Smaller, faster models often suffice for simple schemas.
Monitor Trends
Check the Cost Over Time chart weekly. Sudden spikes may indicate misconfigured batch jobs or unexpected retry loops.
Right-Size Schemas
The Performance by Schema Property Count table shows how cost scales with schema size. Remove unnecessary properties to reduce per-enrichment cost.
Use Caching Models
Models with prompt caching (like Anthropic) reduce costs for repeated enrichments with the same schema. Token Usage cards show cached token savings.