Skip to main content
Home > “Admin Menu” > “Dashboard”

​
Dashboard

The dashboard provides administrators with a central overview of usage and costs in VARIOS AI.
At a glance, you can monitor consumption, cost trends, and usage patterns. Use the period selector in the top left to set the analysis period. All metrics and charts refer to the selected period.

​
Overview of Key Metrics

The top area displays the most important metrics as tiles:
  • Number of Requests
    Total number of requests in the selected period.
  • Total Costs
    Total costs for the selected period in USD.
  • Average per User
    Average costs per active user in the selected period.

​
Chart: Overview

A line chart shows the cost trend in the selected period.
This makes it easy to identify peak loads, trends, and unusual usage patterns.

​
Cost Breakdown

Costs are broken down in multiple panels. Each panel has a search field and pagination.
Costs are based on an extrapolation from transmitted tokens. Deviations from the actual costs billed by the model provider are possible.
User statistics can be enabled or disabled via Admin Menu > Settings > Privacy.

​
Data export

Using the “Export data” function, all usage and cost data can be exported as a CSV file.
This enables further processing in external controlling or reporting tools.
No personal data is included in the export.
Warning: Costs are based on an extrapolation from transmitted tokens.
Deviations from the actual costs billed by the model provider are possible.

​
Included data fields

Each row contains one cost item. A request to an AI model creates one row each for non-cached input, cache reads, cache writes and output, as far as tokens were used for them. The export only contains items with costs greater than 0 and all items of type api (tool and connector calls). The token columns are only filled for the matching type. For systemGeneration, voiceTranscribe and voicePlay, the same row contains both InputTokens and OutputTokens.
  • DateTime: Time of the operation
  • Costs: Cost of the item in USD
  • Type: Type of the item:
    • llmInput: non-cached input tokens
    • llmInputCache: tokens read from the prompt cache
    • llmInputCacheWrite: tokens written to the prompt cache
    • llmOutput: output tokens
    • systemGeneration: system tasks such as chat titles, summaries or service status checks
    • embedding, systemEmbedding: embeddings for knowledge bases and for the service status check
    • voiceTranscribe, voicePlay: speech recognition and speech output
    • api: call of a tool or connector, also without costs
    • variosApi: only appears in older records
  • Tag: Label of the operation, for example chatTitle, chatSummary or variosApi for requests through the API; empty for chat messages
  • Name: Name of the AI model or tool used
  • Assistent: Assistant used to execute the request
  • InputTokens: Number of non-cached input tokens
  • CachedInputTokens: Number of tokens read from the prompt cache
  • OutputTokens: Number of output tokens generated
  • CacheWriteTokens: Number of tokens written to the prompt cache

​
Action

Click “Export CSV” to create the file and automatically start the download.

​
Guide

1

Select period

Use the period selector in the top left to set the desired analysis period. All metrics and charts refer to this period.
2

Review overall status

Use the top tiles (Requests, Total Costs, Average) to get a quick status overview.
3

Analyze cost development

Use the chart to identify trends and peaks in usage.
4

Evaluate cost distribution

Use the panels to view costs by group, assistant, model, connector, or user.
5

Export data

Export the data for further analysis or for billing purposes.
Best practice:
  • Check regularly (e.g., weekly) whether certain models or user groups are causing disproportionately high costs.
  • Use the cost overview by model to deliberately deploy more cost-effective alternatives (e.g., smaller models).
  • Export the data monthly for controlling and budget planning.
  • Example: Use GPT-5 Response only for complex tasks, while standard requests are handled with more affordable models (e.g., GPT-5 Nano).