Dashboard and filters

The dashboard applies one shared analysis selection to headline metrics, charts, tables, CSV export, and PDF report requests. Provider budgets always use the complete selected month (or the current local month) across all systems and models. Provider selection controls which budget rows are visible. Start with the filter bar, then move from overview cards into the analyses that explain the result.

Filter the dataset

The filter bar supports:

daily, monthly, and yearly aggregation
all-time, 7-day, 30-day, current-month, and current-year presets
a specific month
an inclusive custom start and end date
one or more source systems when additional system-transfer files are present
one or more providers
one or more models

Open More filters for custom dates, systems, providers, and models. Active selections remain visible as removable chips even when these controls are collapsed.

System, provider, and model filters combine with the selected time range. An empty system selection means all available systems. Reset all returns to your saved defaults, not necessarily the factory defaults. Configure those defaults in Settings.

Read the analysis flow

TTDash groups analysis into configurable sections:

headline metrics, cost analysis, and insights
today and current-month summaries
cost, request, and token activity calendars
forecasts, cache ROI, and request quality
provider subscriptions and monthly limits
cost, token, and request analyses
distributions, correlations, and concentration risk
period comparisons and anomaly detection
provider, model, and recent-period tables

The first four metrics show total cost, total tokens, requests, and cache share. More metrics · 8 retains the other eight metrics, their explanations, and exact-value tooltips. These details start expanded on desktop and collapsed on mobile. Cost analysis starts with cost over time and model shares; More cost analyses · 6 reveals the other six cost charts. Once opened, supplementary content retains its selections when collapsed or when usage is imported again.

The previous factory section order updates to this overview-first order. Custom section order and visibility settings remain intact.

Use Import for automatic import, Data → Upload for JSON files, Export for PDF reports and CSV, and More → Delete for data removal. Settings, language, theme, and help remain directly accessible.

General metrics use the selected normalized daily rows. Today, current-month, and forecast views retain the source-system/provider/model selection but use their calendar context independently of the analysis date filter. Comparisons retain prior-period data outside that filter. Forecasts are estimates, not billing statements; compare important totals with your provider’s official billing data.

Understand denominators and estimates

Daily cost averages divide by active days, which contain positive cost, tokens, or requests. Monthly and yearly averages divide by the number of displayed periods. Calendar averages include every day of the comparison window.
Week comparisons match Monday through the selected weekday. Month comparisons match the same elapsed days and clamp both windows to the shorter month. Exact dates and missing daily reports are shown. Seven-day moving averages require seven reported calendar days; they remain unavailable across gaps.
Cache share of all tokens divides cache-read tokens by input, output, cache-write, cache-read, and thinking tokens. The separate input cache share excludes output and thinking.
Per-request ratios use only usage with known request counters. The coverage percentage describes that portion of token volume. Missing counters are excluded from request distributions and correlations, rather than treated as measured zeros. Token-per-request histograms represent averages of periods, not individual requests.
Forecasts use the actual current local month. The current partial day contributes recorded cost but is excluded from the trend model. Missing reports are assumed to have zero cost for estimation and cap confidence at low. Total and provider projections add up.
Cache ROI is a standard-price model estimate for supported direct API models, with explicit cost/token coverage and dated provider sources. Read discounts are reduced by cache-write premiums. Unknown Anthropic cache duration produces a 5-minute/1-hour range; net savings can be negative. Models without reliable prices are excluded. Recorded costs are never repriced.
CSV cost values preserve their numeric precision instead of rounding every row to cents.

The data-quality notice shows the most recent reported date, partial request coverage, unreadable sources, and excluded legacy rows. Source files are not changed while reading them.

Drill into a period

Cost and request time-series points open a period drilldown. Recent-period table rows can be activated with a pointer, Enter, or Space. Inside a drilldown, use left and right arrow keys to move between available periods.

Expanded charts preserve the current selection, statistics, extra analyses, and precise CSV values. Zoom windows use a fixed title and close button with a scrolling analysis area; time-series plots grow with the available viewport. Both forecast charts open the same forecast detail window. Close an expanded view or dialog with Escape; focus returns to the control that opened it.

Sections and charts introduce themselves on their first visible appearance. Chart and meter animations run only while their content is visible, including inside scrolling zoom windows, and pause in a hidden browser tab. Scrolling back keeps revealed content in place. Visible value changes use short transitions. The app’s reduced-motion setting also applies to zoom animations and section navigation.

Customize sections

Open Settings to:

show or hide each dashboard section
reorder the sections
choose light or dark theme
switch between German and English
follow the operating-system motion preference, always reduce motion, or never reduce it
record provider subscription prices and monthly limits
save default date, aggregation, system, provider, and model filters

Settings are stored locally in settings.json and can be exported separately from usage data.

Command palette and keyboard use

Open the command palette with Ctrl+K on Windows/Linux or Cmd+K on macOS. It exposes dashboard actions, filters, section navigation, theme and language changes, settings, and help.

To open help, launch the command palette, search for Help, and select the matching action.

Useful keys:

Key	Behavior
`Escape`	Close the active dialog, expanded chart, or palette
Arrow keys	Navigate grids, tabs, supported tables, and drilldown periods
`Enter` / `Space`	Activate the focused interactive row or control
`1`–`9`	Select one of the first nine visible actions while the palette is open

Understand local state

The dashboard does not send usage data to a hosted TTDash service. It reads and writes through the same-origin local server. Browser preferences are normalized through the server settings contract; usage data is normalized before persistence.

Continue with Exports and backups or use Troubleshooting when a view is empty or incomplete.