Usage Metering

The Usage Metering panel tracks AI spend across providers and models, lets you define budgets, and surfaces alerts when consumption crosses thresholds. It’s the operator dashboard for “where is the money going” — Counsel-style panels and agent-loop runs are typically the dominant cost centers.

This page covers the desktop panel. Mobile/watch/IDE clients don’t have a metering surface today; usage data is daemon-side and observable only via the desktop UI or by reading ~/.vibeui/usage-metering.json directly.

Four tabs

Dashboard

Three KPIs at the top:

• Total Spend — running USD total across every recorded request.
• Tokens Used — M for ≥1M, k for ≥1k, raw count otherwise.
• Requests — count of LLM calls.

Below: a horizontal bar chart of spend per provider, with each bar’s width proportional to the provider’s share of total spend. Empty state: “No provider data yet.”

The dashboard is read-only — usage is recorded automatically as the daemon proxies LLM requests. There’s no manual “log this request” affordance because that would defeat the audit-trail purpose.

Budgets

A budget is (name, limit, period). The panel renders one card per budget with:

• A progress bar (role="progressbar" with aria-valuenow) that turns amber at 70% and red at 90%.
• $X used / $Y limit (Z%) summary line.
• Remove button (with two-click intent — but currently a single click; see roadmap).

Create

The Create Budget form takes:

Submitting calls create_usage_budget and persists to ~/.vibeui/usage-metering.json. The new budget appears immediately above the form.

Delete (was broken — fixed)

Bug fix: until this release, “Remove” only mutated local React state. After a page reload the deleted budget came back. The panel now calls a real delete_usage_budget Tauri command that updates the on-disk record.

If you have a workspace with “ghost budgets” that keep returning, edit ~/.vibeui/usage-metering.json directly to remove them — the new delete path won’t fix entries that pre-existed the bug.

Period semantics

The period field controls how used resets. The reset logic is daemon-side (a per-budget cron-like reset at the period boundary) — the UI only displays the current used value. If a budget shows usage that should have rolled over, check RUST_LOG=vibecody::usage=info for the reset event.

Reports

Two views — by provider and by model. Each row: label, tokens (formatted), request count, cost. Sortable by cost (default) — descending.

If both views are empty, no LLM calls have been recorded yet. This is the most common state on a fresh install before any provider is wired up.

Alerts

Alerts appear when:

Each alert card shows the severity badge, the message, and a timestamp. Dismiss removes the alert from the active list (kept in the file with dismissed: true for audit).

/health declaration

features.usage_metering:

{
  "available": true,
  "transport": "tauri-desktop",
  "store_path": "~/.vibeui/usage-metering.json",
  "budget_periods": ["daily", "weekly", "monthly"]
}

Usage metering has no provider dependency — it observes the daemon’s own request log. available is always true.

Observability

Backend operations emit structured tracing events under vibecody::usage:

RUST_LOG=vibecody::usage=info vibecli serve

Events:

INFO vibecody::usage: usage.budget.create
  id=bg1714900000123 name="Marketing Q2" limit=500.0 period=monthly total_budgets=4

INFO vibecody::usage: usage.budget.delete
  id=bg1714900000123 remaining=3

INFO vibecody::usage: usage.alert.dismiss id=al17149...

Budget names are logged (they’re already operator-facing labels), but per-request prompt content, response tokens, and cost details are NOT. Cost aggregation events live under vibecody::cost (a separate target, not in this list yet).

Accessibility

• Tab strip uses role="tablist" with aria-selected per tab.
• Error banner gains role="alert" aria-live="assertive".
• Budget progress bars are exposed as role="progressbar" with aria-valuenow/aria-valuemin/aria-valuemax and a verbose aria-label carrying name, used, limit, period, and percent.
• Remove buttons gain aria-label with the budget name.
• Alert severity badges use color + text label (severity name visible in addition to badge color), so the panel doesn’t rely on color alone.

Cross-client behaviour

Mobile and IDE clients still incur metered usage when they make LLM calls through the daemon — they just can’t view the metering data. A future surface might add a daily-spend-summary push to mobile, but that doesn’t exist today.

Troubleshooting

“Total spend is $0 but I’ve made requests”

The daemon may not be configured to log to the metering store. Check ~/.vibeui/usage-metering.json exists; if not, the metering hooks aren’t wired into the active provider path. Restart the daemon with RUST_LOG=vibecody::usage=info and watch for usage.request.recorded events on each LLM call.

“Budget shows 100% but no alert appeared”

Alerts are generated at the moment the threshold is crossed, not on a periodic check. If the threshold was crossed before the alert dispatcher was wired, no alert will ever fire for it. Increment usage by 1 cent and watch for the alert.

“Removed budgets keep coming back”

You’re on a build before the delete_usage_budget fix. Edit ~/.vibeui/usage-metering.json and remove the entries from the budgets array directly.

“By Task report is always empty”

That column isn’t wired to a real source yet. The task reportData is [] in the panel; the placeholder exists for a future wiring of agent-task → cost attribution. Don’t read into the empty state.

Related

• Cost router: [docs/cost-router/] (TODO) — the routing layer that picks the cheapest provider for a request given quality constraints
• Source: vibeui/src/components/UsageMeteringPanel.tsx (~310 LOC) · backend vibeui/src-tauri/src/commands.rs (create_usage_budget, delete_usage_budget, dismiss_usage_alert, get_usage_kpis, get_usage_budgets, get_usage_by_provider, get_usage_by_model, get_usage_alerts)