CodexBar 0.24

Providers & Usage

• Windsurf: add provider support with web-session usage fetching and local SQLite-cache fallback (#583). Thanks @Coooolfan!
• Codebuff: add provider support with credit balance tracking, weekly rate-limit usage, API-token settings, and codebuff login credential import (#837). Thanks @anandghegde!
• Copilot: add multi-account support with GitHub OAuth sign-in, account switching, and per-account usage cards (#637). Thanks @ajmccall!
• DeepSeek: add provider support with token-account balance tracking, paid vs. granted credit breakdown, and CLI support (#811). Thanks @willytop8!
• Storage: add an opt-in menu view for local provider storage usage with background scans and copyable path breakdowns (#829). Thanks @fatiheminoge!
• OpenRouter and DeepSeek: show remaining account balances in the menu bar, while preserving OpenRouter's API-key limit metric when explicitly selected (#832). Thanks @giuseppebisemi!
• Claude: add a peak-hours menu-card indicator with countdowns and a provider setting to hide it (#611). Thanks @hello-amed!
• Cost history: show per-model cost details as a compact vertical list when hovering daily bars (#513). Thanks @iam-brain!
• Copilot: support GitHub Enterprise hosts for the device-flow login and usage API paths (#827). Thanks @ramzesenok!
• Alibaba: clarify China-region API-key failures when the console endpoint requires a browser session (#628). Thanks @XWind18!

Fixes

• Codex: time out hung codex app-server RPC reads and cap loading animation runtime so stalled refreshes no longer keep the menu bar redrawing indefinitely (#842, #844). Thanks @hyspacex!
• Codex: make OpenAI dashboard refreshes handle non-English pages, lazy-loaded credits history, timeout retries, and unrelated Skillusage rows (#825). Thanks @xiaoqianWX!
• Cursor: show Enterprise/Team usage from personal caps and shared pools instead of reporting 100% remaining (#813). Thanks @fcamus00!
• Codex: keep same-workspace managed accounts distinct by matching workspace identity with email, so different OpenAI users in one workspace no longer overwrite each other (#796). Thanks @leezhuuuuu!
• Claude: enable Claude and switch to OAuth after a successful login, clear stale selected-provider state when Claude is disabled, and tolerate OAuth payloads that omit the five-hour window (#816, #726). Thanks @pdurlej and @Brandawg93!
• Claude: recognize OAuth subscriptionType before rateLimitTier so Pro accounts with generic Claude Code tiers
  open the subscription usage dashboard correctly (#836, fixes #824). Thanks @shixy96!
• Usage: preserve known reset countdowns when a refresh returns current usage without reset metadata (#427). Thanks @Whoaa512!
• Menu: refresh open usage cards after live data changes so the “Updated” timestamp advances after manual or cadence refreshes (#715). Thanks @cooper-matt!
• Menu: make the global open-menu shortcut behave as a true toggle when the menu is already open, avoiding queued reopens after repeated key presses (#218).
• Menu bar: preserve existing status items and assign stable autosave names so provider icon positions survive provider toggles (#538). Thanks @hxy91819!
• Settings: make the Preferences window 10% wider and taller so dense provider/settings panes have more breathing room.
• CLI releases: publish macOS arm64 and x86_64 CLI tarballs alongside Linux artifacts, with release-workflow smoke tests and docs (#457, #839). Thanks @androidshu and @mondary!
• CLI: query only enabled providers by default when three or more providers are enabled instead of expanding to every registered provider (#830). Thanks @lhoBas!
• CLI: read MiniMax coding-plan tokens from MINIMAX_CODING_API_KEY, accept Alibaba Qwen/DashScope API-key aliases, and avoid duplicate generic JSON error rows after provider failures.
• CLI discovery: prefer known install paths before interactive shell probing so common Claude installs no longer run shell init hooks during binary detection (#775).
• CLI lookup: drain login-shell probe output and terminate spawned process groups so interactive shell helpers cannot leak after path detection (#822, fixes #821). Thanks @LPFchan!
• OpenCode Go: open the workspace-specific usage dashboard when a workspace ID is configured (#667). Thanks @RizaSatya!
• Augment: use the API-provided credits limit when available instead of reconstructing the limit from consumed plus remaining credits (#338). Thanks @bcharleson!
• MiniMax: ignore login strings embedded in scripts when checking web-session pages for signed-out state (#508). Thanks @qipihen!
• Accounts: refresh the selected provider data and open menu after switching token accounts, even while a menu-open refresh is running (#799, fixes #798). Thanks @Zeko369!
• Codex: prefer session turn-context model metadata when calculating local cost history so GPT-5.4 sessions are not bucketed as GPT-5 (#620). Thanks @betive37!
• Codex: stop falling back from app-server RPC to bare CLI TUI during automatic usage refreshes, preventing unexpected OpenAI auth browser tabs.
• Menu/keychain: block delayed test-time menu mutations after teardown and enforce no-UI keychain reads more reliably (#381). Thanks @artuskg!

CodexBar 0.23

Highlights

• Mistral: add provider support with monthly spend tracking, browser-cookie import, manual cookies, and CLI/token-account support (#607). Thanks @welcoMattic!
• Claude: show Designs and Daily Routines usage bars from live Claude OAuth/Web quota data, and restore the Web-mode Sonnet bar (#740). Thanks @AISupplyGuy!
• Cursor: add an Extra usage menu bar metric for on-demand budgets (#789). Thanks @huiye98!
• Usage: add an opt-in confetti celebration when weekly limits reset after active use (#785). Thanks @zats!
• Codex: add GPT-5.5 and GPT-5.5 Pro pricing so local cost scanning recognizes the new models.
• Copilot: show a clearer GitHub Device Flow hint in Settings when the copied device code needs to be pasted into GitHub (#369). Thanks @amoranio!

Fixes

• Droid: preserve Factory session fallbacks, use the current usage endpoint, and clarify browser-login messaging (#792). Thanks @JosephDoUrden for the original stale-session fix!
• Widgets: package App Intents metadata for the widget extension and use configuration defaults so configurable widgets load correctly in WidgetKit (#783). Thanks @ngutman and @vincentyangch!
• Menu: keep merged-menu cards, switcher rows, wrapped status text, and hosted chart submenus aligned with the real AppKit menu width so menus no longer grow oversized or show narrower chart submenus after width changes. Thanks @ngutman!
• Codex: ignore invalid zero-minute subscription history so the utilization submenu no longer shows duplicate Session tabs.
• CLI: report the app bundle version correctly when the bundled helper is launched through a symlink.
• Codex/Claude: clean up cached CLI status probes during app shutdown so codex -s read-only workers are not orphaned after restart.

CodexBar 0.22

Highlights

• Codex: restore OpenAI web dashboard fetching on the new analytics route and tighten hidden WebView reuse/expiry.
• Synthetic: parse live quota payloads for five-hour, weekly, and search limits, including continuous reset/regeneration details (#732). Thanks @baanish!
• Antigravity: restore account/quota probing across newer localhost endpoint/token layouts and retry paths (#727). Thanks @icey-zhang!
• Menu: add standard shortcuts for Refresh, Settings, and Quit while the status menu is open (#737). Thanks @anirudhvee!
• Widgets: migrate app-group sharing to the Team-ID-prefixed container and carry widget state across the move (#701). Thanks @ngutman!

Providers & Usage

• Synthetic: parse live five-hour, weekly, and search quota payloads, including continuous reset/regeneration details (#732). Thanks @baanish!
• Antigravity: restore localhost probing with async TLS challenge handling, extension-token fallback, and best-effort port selection (#727). Thanks @icey-zhang!
• Gemini: discover OAuth config in fnm/Homebrew/bundled CLI layouts so expired-token refresh keeps working (#723). Thanks @Leechael!
• Copilot: open the complete device-login verification URL when available so the browser flow carries the user code (#739). Thanks @skhe!
• Alibaba: update the China mainland Coding Plan endpoint and browser-cookie domain while keeping older domains as fallbacks (#712). Thanks @hezhongtang!
• Codex: restore OpenAI web dashboard fetching on the new analytics route and tighten hidden WebView reuse/expiry. @ratulsarna

Menu & Settings

• Menu: show and handle standard shortcuts for Refresh (⌘R), Settings (⌘,), and Quit (⌘Q) while the status menu is open (#737). Thanks @anirudhvee!
• Widgets: migrate app-group sharing to the Team-ID-prefixed container and carry widget state across the move (#701). Thanks @ngutman!
• Settings: fix provider-sidebar clipping on macOS Tahoe and resize the Preferences window when switching tabs (#580). Thanks @chadneal!

Fixes

• Keychain cache: preserve cached credentials when macOS temporarily denies keychain UI after wake, avoiding repeated prompts (#594). Thanks @josepe98!

CodexBar 0.21

Highlights

• Abacus AI: add a new provider for ChatLLM and RouteLLM credit tracking with browser-cookie import, manual-cookie support, and monthly pace rendering. Thanks @ChrisGVE!
• Codex: recognize the new Pro $100 plan in OAuth, OpenAI web, menu, and CLI rendering, and preserve CLI fallback when partial OAuth payloads lose the 5-hour session lane (#691, #709). Thanks @ImLukeF!
• Codex: make OpenAI web extras opt-in for fresh installs, preserve working legacy setups on upgrade, add an OpenAI web battery-saver toggle, and keep account-scoped dashboard state aligned during refreshes and account switches (#529). Thanks @cbrane!
• Codex: fix local cost scanner overcounting and cross-day undercounting across forked sessions, cold-cache refreshes, and sessions-root changes (#698). Thanks @xx205!
• z.ai: preserve weekly and 5-hour token quotas together, surface the 5-hour lane correctly across the menu/menu bar, and add regression coverage (#662). Thanks to @takumi3488 for the original fix and investigation.
• Cursor: fix a crash in the usage fetch path and add regression coverage (#663). Thanks @anirudhvee for the report and validation!
• Antigravity: restore account and quota probing across newer localhost endpoint/token layouts and API-level retry failures (#693, fixes #692). Thanks @anirudhvee!
• Menu bar: fix missing icons on affected macOS 26 systems by avoiding RenderBox-triggering SwiftUI effects (#677). Thanks @andrzejchm!
• Battery / refresh: cut menu redraw churn, skip background work for unavailable providers, and reuse cached OpenAI web views more efficiently (#708).
• Claude: add Opus 4.7 pricing so local cost scanning and cost breakdowns recognize the new model. Thanks @knivram!
• Codex: add Microsoft Edge as a browser-cookie import option for the Codex provider while preserving the contributor-branch workflow from the original PR (#694). Thanks @Astro-Han!

Providers & Usage

• Abacus AI: add provider support for ChatLLM and RouteLLM monthly compute-credit tracking with cookie import, manual cookie headers, timeout/browser-detection threading, optional billing fallback, and hardened cached-session retry behavior. Thanks @ChrisGVE!
• Codex: render the new Pro $100 plan consistently across OAuth, OpenAI web, menu, and CLI surfaces, tolerate newer Codex OAuth payload variants like prolite, and only fall back to the CLI in auto mode when OAuth decode damage actually drops the session lane (#691, #709).
• Codex: make OpenAI web extras opt-in by default, preserve legacy implicit-auto cookie setups during upgrade inference, add battery-saver gating for non-forced dashboard refreshes, and preserve provider/dashboard state for enabled providers that are temporarily unavailable.
• Cost: tighten the local Codex cost scanner around fork inheritance, cold-cache discovery, incremental parsing, and sessions-root changes so replayed sessions no longer overcount or slip usage across day boundaries (#698). Thanks @xx205!
• z.ai: preserve both weekly and 5-hour token quotas, keep the existing 2-limit behavior unchanged, and render the 5-hour quota as a tertiary row in provider snapshots and CLI/menu cards (#662). Credit to @takumi3488 for the original fix and investigation.
• Cursor: fix the usage fetch path so failed or cancelled requests no longer crash, and add Linux build and regression test coverage fixes (#663).
• Antigravity: try both language-server and extension-server endpoint/token combinations, retry after API-level errors, scope insecure localhost trust handling to loopback hosts, and restore local quota/account probing on newer Antigravity builds (#693, fixes #692). Thanks @anirudhvee!
• Antigravity: prefer userTier.name over generic plan info when rendering the account plan so Google AI Ultra and similar tiers show their real subscription name, while still falling back cleanly when the tier label is absent or blank (#303). Thanks @zacklavin11!
• Ollama: recognize __Secure-session cookies during manual cookie entry and browser-cookie import so authenticated usage fetching continues to work with the newer cookie name (#707). Thanks @anirudhvee!
• OpenCode: enable weekly pace visualization for the app and CLI so weekly bars show reserve percentage, expected-usage markers, and "Lasts until reset" details like Codex and Claude (#639). Thanks @zachary!
• Refresh pipeline: skip background work for unavailable providers, clear stale cached state, and show explicit unavailable messages (#708).
• Codex: support Microsoft Edge in browser-cookie import for the Codex provider while keeping the contributor branch untouched in the superseding integration path (#694). Thanks @Astro-Han!
• OpenCode / OpenCode Go: treat serialized _server auth/account-context failures as invalid credentials so cached browser cookies are cleared and retried instead of surfacing a misleading HTTP 500.
• OpenAI web: keep cached WebViews across same-account refreshes and clean them up only when accounts or providers go stale (#708).
• Claude: add Opus 4.7 pricing so local cost usage and breakdowns price the new model correctly. Thanks @knivram!
• Claude: broaden CLI binary lookup to native installer paths (#731). Thanks @dingtang2008!

Menu & Settings

• Menu bar: fix missing icons on affected macOS 26 systems by replacing RenderBox-triggering material/offscreen SwiftUI effects in the provider sidebar and highlighted progress bar (#677). Thanks @andrzejchm!
• z.ai: fix menu bar selection when both weekly and 5-hour quotas are present (#662).
• Menu bar: avoid redundant merged-icon redraws and make hosted chart submenus load lazily without losing provider context (#708).
• Merged menu: when Overview is selected, keep the merged menu bar icon aligned with the first Overview provider in configured order, even while that provider is still loading (#724). Thanks @anirudhvee!
• Codex: add an OpenAI web battery-saver toggle, keep manual refresh available when battery saver is on, and hide OpenAI web submenus when web extras are disabled.

Development & Tooling

• Diagnostics: add lightweight battery instrumentation for menu updates and refresh work (#708).
• Build script: make CodexBar-owned ad-hoc keychain cleanup opt-in with --clear-adhoc-keychain, and extend the explicit reset path to clear both com.steipete.CodexBar and com.steipete.codexbar.cache. Thanks @magnaprog!

CodexBar 0.20

Highlights

• Codex: switch between system accounts/profiles without manually logging out and back in. @ratulsarna
• Add Perplexity provider support with recurring, bonus, and purchased-credit tracking, Pro/Max plan detection, browser-cookie auto-import, and manual-cookie fallback (#449). Thanks @BeelixGit!
• Add OpenCode Go as a separate provider with 5-hour, weekly, and monthly web usage tracking, widget integration, and browser-cookie support.
• Claude: fix token and cost inflation caused by cross-file double counting of subagent JSONL logs, fix streaming chunk deduplication, and add claude-sonnet-4-6 pricing. Thanks @enzonaute for the investigation!
• Cost history: merge supported pi session usage into Codex/Claude provider history (#653). Thanks @ngutman!

Providers & Usage

• Perplexity: add recurring, bonus, and purchased-credit tracking; plan detection for Pro/Max; browser-cookie auto-import; and manual-cookie fallback (#449). Thanks @BeelixGit!
• OpenCode Go: add a dedicated provider, parse live authenticated workspace Go usage from the web app, keep monthly optional and honor workspace env overrides.
• Codex: add workspace attribution for account labels and same-email multi-workspace accounts.
• Codex: reconcile live-system and managed accounts by canonical identity, preserve account-scoped usage/history/dashboard state, allow OAuth CLI fallback, and tighten OpenAI web ownership gating so quota and credits only attach to the matching account. Thanks @monterrr and @Rag30 for the initial effort and ideas!
• Codex: normalize weekly-only rate limits across OAuth and CLI/RPC so free-plan accounts render as Weekly instead of a fake Session, preserve unknown single-window payloads in the primary lane, hide the empty Session lane in widgets, and accept weekly-only Codex CLI /status/RPC data without failing. @ratulsarna
• Codex: refactor the provider end to end into clearer components and better division of responsibilities.
• OpenCode: preserve product separation between Zen and Go, improve null/unsupported usage handling, and harden cookie/domain behavior for authenticated web fetches.
• Cost history: merge supported pi session usage into Codex/Claude provider history (#653). Thanks @ngutman!

Menu & Settings

• Codex: add UI for switching the system-level Codex account and promoting a managed account into the live system slot.
• Codex: hide display-only OpenAI web extras in widgets and fix buy-credits / credits-only presentation regressions.
• Claude: enable “Avoid Keychain prompts” by default, remove the experimental label, and preserve user-action cooldown clearing plus startup bootstrap when Security.framework fallback is still needed.
• Fix alignment of menu chart hover coordinates on macOS. Thanks @cuidong233!

CodexBar 0.20.0-beta.1

Highlights

• Add basic multi-account support to Codex. Thanks @monterrr and @Rag30 for the initial effort and ideas!
• Add Perplexity provider with recurring, bonus, and purchased-credit tracking; plan detection (Pro/Max); and browser-cookie auto-import with manual-cookie fallback (#449). Thanks @BeelixGit!

Providers & Usage

• Add the foundation for multi-account support to Codex and basic UX for adding and switching accounts. @ratulsarna
• Codex: normalize weekly-only rate limits across OAuth and CLI/RPC so free-plan accounts render as Weekly instead of a fake Session, preserve unknown single-window payloads in the primary lane, hide the empty Session lane in widgets, and accept weekly-only Codex CLI /status/RPC data without failing. @ratulsarna
• Perplexity: add provider support with credit tracking for recurring (monthly), bonus (promotional), and purchased on-demand credits; plan detection (Pro/Max); and browser-cookie auto-import with manual-cookie fallback (#449). Thanks @BeelixGit!

Menu & Settings

• Fix alignment of menu chart hover coordinates on macOS. Thanks @cuidong233!

CodexBar 0.19.0

Highlights

• Add Alibaba Coding Plan provider with region-aware quota fetching, widget integration, and browser-cookie import defaults (#574).
• Align Cursor usage with the dashboard's Total/Auto/API lanes. (#587). Thanks @Rag30!
• Add subscription utilization history chart to the menu with DST-safe data point identification (#589). Thanks @maxceem!
• Refactor the Claude provider end to end into clearer, better-tested components while preserving behavior (#494). @ratulsarna
• Add reset time display for Codex code review limits (#581). Thanks @Q1CHENL!
• Add per-model token counts to cost history (#546). Thanks @iam-brain!
• Fix Antigravity model selection to use stable model-family matching for Claude, Gemini Pro, and Gemini Flash, and preserve fallback lane visibility in the menu bar and icon (#590). Thanks @skainguyen1412!
• Add GPT-5.4 mini and nano pricing (#561). Thanks @iam-brain!

Providers & Usage

• Alibaba: add Coding Plan provider support with region-aware web/API quota fetching, widget integration, and browser-cookie import defaults (#574).
• Cursor: trust dashboard percent fields for Total/Auto/API usage, preserve on-demand remaining fallback views, and keep scanning imported browser-cookie candidates until a working Cursor session is found (#587, supersedes #579). Thanks @Rag30!
• Claude: refactor the provider end to end into clearer components, with baseline docs and expanded tests to lock down behavior (#494).
• Codex: show reset times for code review limits, including Core review reset parsing support (#581). Thanks @Q1CHENL!
• Cost history: add per-model token counts so token usage is broken out by model (#546). Thanks @iam-brain!
• Antigravity: replace label-order guessing with stable model-family selection for Claude, Gemini Pro, and Gemini Flash; fix mapping for Claude thinking models and placeholder model IDs; preserve fallback lane visibility in the menu bar and icon when only fallback lanes exist (#590). Thanks @skainguyen1412!
• Kimi: tolerate API responses without resetTime so usage decoding no longer fails on sparse payloads.
• Codex: add GPT-5.4 mini and nano pricing (#561). Thanks @iam-brain!

Menu & Settings

• Menu: add subscription utilization history chart with DST-safe chart point identifiers and per-provider plan utilization tracking (#589). Thanks @maxceem!
• Menu bar: in Both display mode, fall back to percent when pace data is unavailable so text stays visible for providers without pace metrics (#527). Thanks @Astro-Han!
• Settings: persist the resolved refresh cadence default to UserDefaults on first launch and repair invalid stored values so the setting stays normalized across relaunches (#519). Thanks @Astro-Han!
• Menu: wrap long status blurbs and preserve wrapped titles for multiline entries (#543). Thanks @zkforge!

CodexBar 0.18.0

Highlights

• Add Kilo provider support with API/CLI source modes, widget integration, and pass/credit handling (#454). Built on work by @coreh.
• Add Ollama provider, including token-account support in Settings and CLI (#380). Thanks @CryptoSageSnr!
• Add OpenRouter provider for credit-based usage tracking (#396). Thanks @chountalas!
• Add Codex historical pace with risk forecasting, backfill, and zero-usage-day handling (#482, supersedes #438). Thanks @tristanmanchester!
• Add a merged-menu Overview tab with configurable providers and row-to-provider navigation (#416). @ratulsarna
• Add an experimental option to suppress Claude Keychain prompts (#388).
• Reduce CPU/energy regressions and JSONL scanner overhead in Codex/web usage paths (#402, #392). Thanks @bald-ai and @asonawalla!

Providers & Usage

• Codex: add historical pace risk forecasting and backfill, gate pace computation by display mode, and handle zero-usage days in historical data (#482, supersedes #438). Thanks @tristanmanchester!
• Kilo: add provider support with source-mode fallback, clearer credential/login guidance, auto top-up activity labeling, zero-balance credit handling, and pass parsing/menu rendering (#454). Thanks @coreh!
• Ollama: add provider support with token-account support in app/CLI, Chrome-default auto cookie import, and manual-cookie mode (#380). Thanks @CryptoSageSnr!
• OpenRouter: add provider support with credit tracking, key-quota popup support, token-account labels, fallback status icons, and updated icon/color (#396). Thanks @chountalas!
• Gemini: show separate Pro, Flash, and Flash Lite meters by splitting Gemini CLI quota buckets for gemini-2.5-flash and gemini-2.5-flash-lite (#496). Thanks @aladh
• Codex: in percent display mode with "show remaining," show remaining credits in the menu bar when session or weekly usage is exhausted (#336). Thanks @teron131!
• Claude: surface rate-limit errors from the CLI /usage probe with a user-friendly message, and harden "Failed to load usage data" matching against whitespace-collapsed output.
• Claude: restore weekly/Sonnet reset parsing from whitespace-collapsed CLI /usage output so reset times and pace details still appear after CLI fallback.
• Claude: fix extra-usage double conversion so OAuth/Web values stay on a single normalization path (#472, supersedes #463). Thanks @Priyans-hu!
• Claude: remove root-directory mtime short-circuiting in cost scanning so new session logs inside existing ~/.claude/projects/* folders are discovered reliably (#462, fixes #411). Thanks @Priyans-hu!
• Copilot: harden free-plan quota parsing and fallback behavior by treating underdetermined values as unknown, preserving missing metadata as nil (#432, supersedes #393). Thanks @emanuelst!
• OpenCode: treat explicit null subscription responses as missing usage data, skip POST fallback, and return a clearer workspace-specific error (#412).
• OpenCode: surface clearer HTTP errors. Thanks @SalimBinYousuf1!
• Codex: preserve exact GPT-5 model IDs in local cost history, add GPT-5.4 pricing, and label zero-cost gpt-5.3-codex-spark sessions as "Research Preview" in cost breakdowns (#511). Thanks @iam-brain!
• Augment: prevent refresh stalls when auggie account status hangs by replacing unbounded CLI waits with timed subprocess execution and fallback handling (#481). Thanks @bryant24hao!
• Update Kiro parsing for kiro-cli 1.24+ / Q Developer formats and non-managed plan handling (#288). Thanks @kilhyeonjun!
• Kimi: in automatic metric mode, prioritize the 5-hour rate-limit window for menu bar and merged highest-usage calculations (#390). Thanks @ajaxjiang96!
• Browser cookie import: match Gecko *.default* profile directories case-insensitively so Firefox/Zen cookie detection works with uppercase .Default directories (#422). Thanks @bald-ai!
• MiniMax: make both Settings "Open Coding Plan" actions region-aware so China mainland selection opens platform.minimaxi.com instead of the global domain (#426, fixes #378). Thanks @bald-ai!
• Menu: rebuild the merged provider switcher when “Show usage as used” changes so switcher progress updates immediately (#306). Thanks @Flohhhhh!
• Warp: update API key setup guidance.
• Claude: update the "not installed" help link to the current Claude Code documentation URL (#431). Thanks @skebby11!
• Fix Claude setup message package name (#376). Thanks @daegwang!

Menu & Settings

• Merged menu: keep Merge Icons, the switcher, and Overview tied to user-enabled providers even when some providers are temporarily unavailable, while defaulting menu content and icon state to an available provider when possible (#525). Thanks @Astro-Han!
• Merged menu: add an Overview switcher tab that shows up to three provider usage rows in provider order (#416).
• Settings: add "Overview tab providers" controls to choose/deselect Overview providers, with persisted selection reconciliation as enabled providers change (#416).
• Menu: hide contextual provider actions while Overview is selected and rebuild switcher state when overview availability changes (#416).

Claude OAuth & Keychain

• Add an experimental Claude OAuth Security-CLI reader path and option in settings.
• Apply stored prompt mode and fallback policy to silent/noninteractive keychain probes.
• Add cooldown for background OAuth keychain retries.
• Disable experimental toggle when keychain access is disabled.
• Use a claude-code/<version> User-Agent for OAuth usage requests instead of a generic identifier.

Performance & Reliability

• Codex/OpenAI web: reduce CPU and energy overhead by shortening failed CLI probe windows, capping web retry timeouts, and using adaptive idle blink scheduling (#402). Thanks @bald-ai!
• Cost usage scanner: optimize JSONL chunk parsing to avoid buffer-front removal overhead on large logs (#392). Thanks @asonawalla!
• TTY runner: fence shutdown registration to avoid launch/shutdown races, isolate process groups before shutdown rejection, and ensure lingering CLI descendants are cleaned up on app termination (#429). Thanks @uraimo!

CodexBar 0.18.0-beta.3

Highlights

• Claude OAuth/keychain flows were reworked across a series of follow-up PRs to reduce prompt storms, stabilize background behavior, surface a setting to control prompt policy and make failure modes deterministic (#245, #305, #308, #309, #364). Thanks @manikv12!
• Claude: harden Claude Code PTY capture for /usage and /status (prompt automation, safer command palette confirmation, partial UTF-8 handling, and parsing guards against status-bar context meters) (#320).
• New provider: Warp (credits + add-on credits) (#352). Thanks @Kathie-yu!
• Provider correctness fixes landed for Cursor plan parsing and MiniMax region routing (#240, #234, #344). Thanks @robinebers and @theglove44!
• Menu bar animation behavior was hardened in merged mode and fallback mode (#283, #291). Thanks @vignesh07 and @Ilakiancs!
• CI/tooling reliability improved via pinned lint tools, deterministic macOS test execution, and PTY timing test stabilization plus Node 24-ready GitHub Actions upgrades (#292, #312, #290).

Claude OAuth & Keychain

• Claude OAuth creds are cached in CodexBar Keychain to reduce repeated prompts.
• Prompts can still appear when Claude OAuth credentials are expired, invalid, or missing and re-auth is required.
• In Auto mode, background refresh keeps prompts suppressed; interactive prompts are limited to user actions (menu open or manual refresh).
• OAuth-only mode remains strict (no silent Web/CLI fallback); Auto mode may do one delegated CLI refresh + one OAuth retry before falling back.
• Preferences now expose a Claude Keychain prompt policy (Never / Only on user action / Always allow prompts) under Providers → Claude; if global Keychain access is disabled in Advanced, this control remains visible but inactive.

Provider & Usage Fixes

• Warp: add Warp provider support (credits + add-on credits), configurable via Settings or WARP_API_KEY/WARP_TOKEN (#352). Thanks @Kathie-yu!
• Cursor: compute usage against plan.limit rather than breakdown.total to avoid incorrect limit interpretation (#240). Thanks @robinebers!
• MiniMax: correct API region URL selection to route requests to the expected regional endpoint (#234). Thanks @theglove44!
• MiniMax: always show the API region picker and retry the China endpoint when the global host rejects the token to avoid upgrade regressions for users without a persisted region (#344). Thanks @apoorvdarshan!
• Claude: add Opus 4.6 pricing so token cost scanning tracks USD consumed correctly (#348). Thanks @arandaschimpf!
• z.ai: handle quota responses with missing token-limit fields, avoid incorrect used-percent calculations, and harden empty-response behavior with safer logging (#346). Thanks @MohamedMohana and @halilertekin!
• z.ai: fix provider visibility in the menu when enabled with token-account credentials (availability now considers the effective fetch environment).
• Amp: detect login redirects during usage fetch and fail fast when the session is invalid (#339). Thanks @JosephDoUrden!
• Resource loading: fix app bundle lookup path to avoid "could not load resource bundle" startup failures (#223). Thanks @validatedev!
• OpenAI Web dashboard: keep WebView instances cached for reuse to reduce repeated network fetch overhead; tests were updated to avoid network-dependent flakes (#284). Thanks @vignesh07!
• Token-account precedence: selected token account env injection now correctly overrides provider config apiKey values in app and CLI environments. Thanks @arvindcr4!
• Claude: make Claude CLI probing more resilient by scoping auto-input to the active subcommand and trimming to the latest Usage panel before parsing to avoid false matches from earlier screen fragments (#320).

Menu Bar & UI Behavior

• Prevent fallback-provider loading animation loops (battery/CPU drain when no providers are enabled) (#283). Thanks @vignesh07!
• Prevent status overlay rendering for disabled providers while in merged mode (#291). Thanks @Ilakiancs!

CI, Tooling & Test Stability

• Pin SwiftFormat/SwiftLint versions and harden lint installer behavior (version drift + temp-file leak fixes) (#292).
• Use more deterministic macOS CI test settings (including non-parallel paths where needed) and align runner/toolchain behavior for stability (#292).
• Stabilize PTY command timing tests to reduce CI flakiness (#312).
• Upgrade actions/checkout to v6 and actions/github-script to v8 for Node 24 compatibility in upstream-monitor.yml (#290). Thanks @salmanmkc!
• Tests: add TaskLocal-based keychain/cache overrides so keychain gating and KeychainCacheStore test stores do not leak across concurrent test execution (#320).

Docs & Maintenance

• Update docs for Claude data fetch behavior and keychain troubleshooting notes.
• Update MIT license year.

CodexBar 0.18.0-beta.2

Highlights

• OpenAI web dashboard refresh cadence now follows 5× the base refresh interval.
• OpenAI web dashboard WebView is torn down after each scrape to reduce idle CPU.
• Codex settings now include a toggle to disable OpenAI web extras.

Providers

• Providers: add Dia browser support across cookie import and profile detection (#209). Thanks @validatedev!
• Codex: include archived session logs in local token cost scanning and dedupe by session id.
• Claude: harden CLI /usage parsing and avoid ANTHROPIC_* env interference during probes.

Menu & Menu Bar

• Menu: opening OpenAI web submenus triggers a refresh when the data is stale.
• Menu: fix usage line labels to honor “Show usage as used”.
• Debug: add a toggle to keep Codex/Claude CLI sessions alive between probes.
• Debug: add a button to reset CLI probe sessions.
• App icon: use the classic icon on macOS 15 and earlier while keeping Liquid Glass for macOS 26+ (#178). Thanks @zerone0x!