2.0.0

What's new in 2.0

Cost correctness

• Cache writes billed by TTL — Anthropic charges 1.25× base input for a 5-minute cache entry but 2× for a 1-hour one. TokenMeter 1.x applied the cheaper rate to everything; this understated cost by up to ~6% in cache-heavy months.
• Server tool calls are billed — web_search_requests is now charged per request, not per token.
• Bundled price table ships current — the offline fallback pricing table is regenerated at release time so new models are never silently unpriced.

Plan value

Settings → Plan. Tell TokenMeter what you pay (Claude Pro, Max 5×/20×, ChatGPT Plus/Pro, pay-as-you-go, or custom) and the Overview reframes cost as a return-on-plan multiple, anchored to your renewal day.

Budgets & notifications

Settings → Budgets. Ceilings on the 5-hour block, day, week, or billing period, with alerts at 50/80/100% and anti-spam de-duplication per window instance.

Prompt-cache analysis

The Overview panel shows your cache hit rate and what caching is actually saving you against the same traffic billed with no cache tiers — the most actionable number in the app.

Attribution

History now breaks usage down by git branch, session, subagent, and skill — not just by model.

Export

CSV/JSON export for any period, with per-event cost and full attribution. CSV output is formula-injection-safe.

Menu-bar & launch behavior

Live status-item readout (tokens, cost, block progress, or plan multiple). The app no longer force-opens its window on every launch.

Update checks

Settings → General checks GitHub Releases for a newer version — no auto-download, no auto-install, no telemetry.

Performance

Persisted read offsets and day-bucketed rollups: cold start ~34s → ~2s, refresh 349ms → 0.098ms on a 343 MB / 25,000-event log tree.

Settings navigation

Settings is now a sidebar list instead of an unlabeled icon strip.

See the README for full details.