A new icon, and more accurate numbers TokenWatch has a new flat icon, sharp even at the smallest sizes in Finder and Spotlight, and its own small mark beside your reading in the menu bar. Counting and pricing fixes mean some Claude Code and Codex figures will change. Nothing was lost. Also in this release: Usage History opens faster and the popover steps aside for it; history charts give every calendar day its own bar; models of one family get their own shades; CSV dates follow your time zone; quiet hours that start or end at midnight are kept; the Admin API cost report is read in the format the API sends; VoiceOver names the menu bar item and Settings switches; and Turkish capitals and a few clipped labels are fixed.

Claude Code's hours, and a few things that were in the way The heatmap has drawn Codex's hours since the day it shipped. Claude Code's were being collected the whole time and never shown. They are on that tab now, read from your own ~/.claude, on your own clock. . Small widgets show when the window resets. The reset sat behind the cache ratio, and that ratio exists the moment there is any Codex history to read, so almost nobody ever saw it . The range you pick in one window stays in that window. A week chosen in the popover used to rewrite the detached window behind it, and CSV export wrote whichever range the popover was holding rather than the one on screen . Accent text is darker where it is small. The ETA caption and the projection line are the two lines that ask you to act, and they were the two hardest to read . One banner at a time instead of three Also in this release: notifications are asked for when you connect an agent rather than at first launch, and the alerts pane says so when they are switched off; a failed organisation list shows what went wrong instead of nothing at all; the locked preview is drawn from your own last readings rather than invented ones; and a count of conversations is no longer called sessions, which is the name the limit window already had.

One Claude tab, and what it costs claude.ai and Claude Code spend one 5-hour limit. Instead of two tabs reading the same window differently, there is now one tab and one dial. . That tab shows what Claude Code put into the window you are in, and what the day has cost so far . The info key opens the rest: what caching took off this window, and which models spent it . The weekly limit names the day it runs out at your current pace . A window running well above your usual ones says so, once there are enough behind it to compare One rebuilt interface . The dial is now the only interface: session, weekly and Codex windows on one instrument. The previous layout has been removed . If you had switched back to that layout, this update carries you over to the dial . Cache read and write tokens are on screen for the first time: often 95% of your traffic Your numbers will look smaller. They were wrong before, not now. Four counting bugs inflated tokens and cost. Claude Code split its lifetime total with a ratio that ignored cache reads, which put Input and Output in billions instead of millions. Opus has been priced 5/25 rather than 15/75 since 4.5, and fable-5 was billed at Sonnet's rate. Codex counted a resumed session's history a second time and added reasoning tokens on top of the output that already contains them. Nothing was deleted. The same work is measured correctly now. Codex windows you can trust . CLI builds that publish only a weekly window no longer show 0% or "..." . An expired window resets instead of repeating a days-old percentage, so no false alerts . The weekly countdown now reads 6d 3h instead of 147h Your calendar, not UTC . Work after midnight lands on the right day, and busiest hour is correct . ETA stops at the window reset instead of predicting a limit you never hit Also in this release: a failed read shows as no data rather than 0%, one colour threshold across menu bar, dial and widgets, a Fable weekly limit in widgets, corrected 7-day and 14-day ranges, the month's API cost moved to the claude.ai tab, a way back in when your session dies, open windows now follow a language change, and Turkish for the remaining labels, dates and purchase errors.

Various UI fixes and improvements

Codex CLI support . Local parser for ~/.codex/sessions, no API key required . Live 5h cache window utilisation, weekly limits, and reset countdown . Cache-hit ratio surfaced as a headline stat (typical: 93%+) . Per-day sessions and per-project token breakdown Dual-agent menu bar . New "Both" mode shows Claude and Codex side by side as branded pills . Per-agent gauge colour, so the safer agent stays green while the hot one warns . Picker for Claude only, Codex only, or both Popover dual hero . Segmented agent switcher at the top of the popover . Hide either agent in Settings if you only use one (tabs disappear too) Widget agent picker . Edit any widget to monitor Claude, Codex, or both . Small "Both" shows whichever agent is closest to its limit . Medium and Large "Both" show two centred gauges with stats below . Falls back to Claude if Codex is not connected yet Detailed History . New Codex tab with model breakdown, sessions, and top projects . Date range filter shared across Claude Code and Codex tabs Performance . Codex parser uses the same per-file mtime cache pattern as Claude Code . Cold parse around 5s on a 500MB session history, warm reads under 500ms

What's New in 1.2.0 TokenWatch Pro . New 7-day free trial for first-time users . One-time $1.99 unlock, no subscription . Customers from earlier paid versions keep Pro automatically Real-time Claude Code analytics . Live token, session, and cost data parsed from your local Claude Code logs . No more stale "X days old" data; everything updates as you work . 7d / 14d / 30d / All filter to slice by time window . Per-version model breakdown (Opus 4.6, Opus 4.7, Sonnet 4.6, Haiku 4.5) Desktop widgets . Small, Medium, and Large widgets for your desktop and Notification Center . Color-coded session gauge and weekly limit progress at a glance . Tap any widget to jump straight into the app Cleaner menu bar . New "Icon Only" mode shrinks the status item to a single colored dot . Settings explains what to do if the icon gets clipped behind app menus . Cmd-drag tip in onboarding for placing the icon where you want it Faster everywhere . Claude Code parser is up to 10x faster on first launch . Subsequent reads are near-instant thanks to a per-file cache . Filter switches and history window open without freezing Universal Binary . Native arm64 and x86_64 support, no Rosetta on Intel Macs

- Free 14-day trial, try everything before you buy - One-time unlock: $1.99 for lifetime access - Claude Code analytics: token breakdown by model, sessions & activity - Desktop widgets (Small, Medium, Large) usage at a glance - 30-day history with charts, heatmap and model breakdown - Faster startup: instant login restore on Mac restart - Stale data indicator for Claude Code cache - Bug fixes and performance improvements

What's New v1.1.1 Desktop Widgets - Small, Medium, and Large widgets for your desktop - Shows "App not running" overlay when data is stale (5 min) - Tap widget to relaunch the app instantly Menu Bar Display Modes - Full: icon + percentage + ETA - Compact: icon + percentage only - Text Only: percentage + ETA without icon - Compact Text: percentage only, no icon perfect for notched MacBooks Intel Mac Support - Native Universal Binary (arm64 + x86_64) - No more Rosetta translation overhead - Optimized heatmap rendering for Intel GPUs

What's New in v1.1.0 Desktop Widgets • Small, Medium, and Large WidgetKit widgets for your desktop • Live session gauge, weekly limits, and 7-day usage chart • Auto-updates and follows your app language Usage History & Analytics • New detached window with full usage analytics • 7-day / 14-day / 30-day date range filter • Activity heatmap — see when you use Claude most • Tap any bar to see day details: peak, average, snapshots, top model • Highest usage days ranking • Export your full history as CSV Redesigned Dashboard • Bigger session percentage with burn rate pill • 24-hour sparkline showing usage trend • All Models row in Weekly Limits Model Breakdown • Donut chart with per-model usage share • Version-based colors: Opus 4.5/4.6, Sonnet 4.5/4.6, Haiku Claude Code Analytics (Admin API key required) • See which models your Claude Code sessions consume • Token usage, session count, and cost breakdown • Date-range aware filtering Improved Experience • New guided onboarding with step-by-step checklist • Session stays active after restart — no re-login needed • Admin API error messages shown inline • Bug fixes and performance improvements