Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.toktra.dev/llms.txt

Use this file to discover all available pages before exploring further.

The Toktra usage dashboard gives you a complete picture of how your organization uses AI — across every provider, model, department, and user. This guide explains how to read the data, apply filters, export records, and act on efficiency insights.

​
The Overview page

The Overview page is your starting point after logging in. It shows four summary tiles for the last 30 days:
TileWhat it shows
Enrolled DevicesNumber of devices with active agent certificates
Active UsersNumber of users with at least one usage event
Total Tokens (30d)Sum of estimated input + output tokens across all providers
ProtectedWhether active monitoring is running
The Recent Activity panel below summarizes the number of provider usage records in the selected window.

​
The Usage page

Navigate to Usage in the sidebar to open the detailed usage view.

​
Time range

Use the 7d / 30d / 90d buttons in the top-right corner to change the analysis window. Charts and tables update immediately.

​
Tokens by provider

The Tokens by Provider bar chart breaks input and output token counts by provider: OpenAI, Anthropic, Azure OpenAI, and GCP Vertex AI. Use this view to understand which services your organization relies on most.

​
Top users by token volume

The Top Users table ranks users by total tokens consumed in the selected period. A progress bar shows each user’s share relative to the highest consumer in the period.

​
Filtering usage

Use the user filter on the Usage page, or pass user_id as a query parameter to GET /v1/usage to scope the view to a single employee.
Filter by model name — for example, gpt-4o, claude-3-5-sonnet, or gemini-1.5-pro — to understand which model tiers are in use across your org.
Department-level filtering is available when department budgets are configured. Navigate to Budget → Department Hierarchy for per-department usage, or add a department_id filter on the usage API endpoint.
Set a custom date range using start_date and end_date query parameters (format: YYYY-MM-DD) on the API, or use the time range selector in the dashboard.

​
Understanding token counts

Toktra reports two types of token counts:
TypeDescription
EstimatedCounts produced by the on-device token estimator. Available in real time as usage happens.
ReconciledAuthoritative counts pulled from provider admin APIs (OpenAI, Anthropic, Azure, GCP). Matched against device telemetry and used for billing, budget enforcement, and efficiency scoring.
Reconciled counts are the source of truth for billing and budgets. Estimated counts power real-time dashboards because they are available immediately, before the provider polling cycle completes.

​
Provider polling

Toktra polls provider admin APIs to fetch authoritative token counts for your organization:
ProviderSource
OpenAIOpenAI Admin API (organization usage)
AnthropicAnthropic API (Claude usage metrics)
Azure OpenAIAzure ARM API (resource usage)
GCP Vertex AICloud Billing / Metrics API
Once a polling cycle completes, reconciled counts are reflected in the dashboard on the next page load.

​
Exporting usage data

Export raw usage records for analysis in BI tools, compliance audits, or finance systems.
1

Request an export

Call POST /v1/exports with the date range, format, and scope:
{
  "date_from": "2026-03-01",
  "date_to": "2026-03-31",
  "format": "csv",
  "scope": "users",
  "filters": {
    "providers": ["openai", "anthropic"]
  }
}
Valid format values: csv, json. Valid scope values: users, teams, models, providers.
2

Poll for completion

The export runs asynchronously. Poll GET /v1/exports/{job_id} until status is ready.
3

Download the file

Call GET /v1/exports/{job_id}/download. The response redirects you to a pre-signed download URL. The link expires after one hour.
Use the filters.model_names field to export only usage from expensive models — for example, gpt-4o or claude-3-opus — when preparing cost review reports.

​
Scheduled reports

Toktra generates and emails usage, budget, and efficiency reports on a schedule you configure. Reports include:
  • Usage summary by provider and model
  • Top users by token volume
  • Budget utilization vs. limits
  • Efficiency scores and coaching highlights
Configure report frequency and recipients at Settings → Scheduled Reports.

​
Efficiency scores and team leaderboards

Toktra computes per-user and per-team efficiency scores from usage patterns. Scores reflect how effectively tokens are being used — users who achieve more productive output with fewer tokens score higher.

​
Performance dashboard

Navigate to Performance to see:
  • Individual scores: Each user’s current efficiency score with a trend indicator.
  • Team leaderboard: Rankings by efficiency within a department. An anonymization toggle lets you share the leaderboard without exposing individual names.
  • Model access tiers: Users are assigned standard, advanced, or premium tiers based on their efficiency score, controlling which models they can access.

​
AI adoption scorecards

Toktra computes AI adoption scorecards nightly per employee, team, and department. Navigate to Scorecards to view:
  • Adoption rate (percentage of eligible employees actively using AI tools)
  • Token consumption trends over time
  • Model diversity (number of distinct models used)
  • Efficiency trend over 30 and 90 days

​
30/90-day forecasting

The Usage Forecast page projects token consumption 30 and 90 days forward based on the last 90 days of reconciled data. Each forecast includes a trend classification:
ClassificationMeaning
AcceleratingUsage growing faster than the baseline trend
SteadyUsage growing in line with the historical trend
DeceleratingUsage growth slowing
DecliningUsage falling month-over-month
Use the forecast view when planning budget periods or capacity for the next quarter.

​
ROI calculator

Navigate to Performance → ROI Calculator to model the cost impact of different model configurations. Input your current token volumes, model mix, and cost per token to see projected savings from shifting workloads to more efficient models or lower-cost providers.