Extension release
UsageMeterv1.4.3
11 changes
- 7 new
- 1 improved
- 3 fixes
- New
Current 5-hour window, whole account — a new card shows how many tokens you burned in the live window, split per model, merged across every device: Claude Code turns from the companion app AND claude.ai turns from this browser draw from the same budget, and until now nothing showed you both halves at once. Pick which model the card highlights in Settings → Meters
- New
Measured ceiling — the "X / ~Y tokens" figure. Anthropic never publishes the real cap, so UsageMeter measures it from observed usage and pools those measurements across accounts on the same Claude plan (Pro / Max 5× / Max 20×), which converges far faster than any single account could. The card always states where its ceiling came from: your own measurements or the pooled one
- Fix
The "Real token telemetry — waiting for the first message" row never resolved for most people. It is gone, replaced by the actual token accounting above: real numbers from real turns instead of a diagnostic that only reported whether numbers were available
- New
Code tab — a dedicated Claude Code panel in the popup: today/7-day token totals, per-model breakdown, estimated API-equivalent cost (Pro) and your paired UsageMeter Link devices with sync freshness, straight from your usagemeter.app account. Not paired yet? The tab walks you through installing the companion
- Fix
Dashboard sync actually syncs now — fresh installs used to ship with sync OFF, silently leaving usagemeter.app dashboards empty. Sync is now ON by default and a one-time migration re-enables installs stuck on the old default; turning it off in Settings afterwards is still respected across updates. Synced data remains usage percentages and reset times only, never conversation content
- New
Real token telemetry — when claude.ai's response stream carries real usage numbers, UsageMeter records the exact input/output tokens of each message (the usage tooltip gains a "Last message: N in / M out (exact)" line) and the token estimator continuously calibrates itself against those real counts. When the stream carries none, everything stays estimate-based exactly as before — Settings → Meters shows which mode is active on your account
- New
UsageMeter Link — a new Windows companion app that tracks your Claude Code terminal usage into the same usagemeter.app dashboard: multi-PC (each machine appears as its own renamable device), history kept server-side beyond Claude Code's ~30-day local purge, estimated API-equivalent cost, one-time PIN pairing. The extension links to it from the Usage view and Settings → Integrations, and the promo hides automatically once a device is paired
- New
Dashboard connection — signed-in users with dashboard sync enabled now also sync their real per-message web token counts (never conversation content); a dashboard shortcut lives in the popup header, and a one-time prompt after sign-in explains the sync trade-off (it stays OFF unless you say yes)
- New
Connection expiry reminder — dashboard sessions last 90 days and can't renew silently. UsageMeter now warns once, 7 days before expiry (one notification per episode, opt-out in Settings → Meters), with a banner and one-click reconnect that heals the session through usagemeter.app
- Fix
"Approximate usage" recorded each observation with the WHOLE conversation length instead of the message's own cost, inflating the estimated caps. Observations now use the real per-message cost (or the measured increment when telemetry is unavailable); previously mis-measured observations are discarded, so the "≈tokens left" figure goes quiet briefly and recalibrates over your next messages — honest beats wrong
- Improved
The "context left" figure now follows the active model's real context window (1M-context models are no longer read against a hardcoded 200k), and the internal token estimator learns your account's real system-prompt overhead instead of assuming a frozen constant