midas
MIDAS 0.48.4 / COMPLETE SYNC

Know what your
AI is worth.

Midas sits in your Mac’s menu bar and shows what your Codex, Claude, Cursor, and Muse Code usage would cost at API rates, how much of your Codex, Claude, and Cursor quota is left, and when it resets. Spend your subscriptions on purpose, not on guesswork.

FREE & OPEN SOURCE · APPLE SILICON · macOS 14+ · SIGNED & NOTARIZED

WORKS WITH THE TOOLS
YOU ALREADY PAY FOR.

50+ PROVIDERS
01 / THE OVERVIEW

Your whole AI stack.
One click away.

Click the menu bar and Midas shows estimated inference spend for the period you choose, then breaks it down by provider. Each connected Codex account gets its own remaining-usage bar, reset time, and banked reset count.

Try the interactive preview
YOUR MENU BAR
$2489:41
Midas

Estimated inference spend (API rates)

$248.60
Sep 1 – 13·3 providers
Codex
$186.40
Weekly remaining72% left
Banked resets2 Work · Unavailable Personal

INTERACTIVE PREVIEW ↗ SWITCH THE PERIOD OR THE FAVORITE PROVIDER

Illustrative values, not real accounts. Usage sharing is optional.

02 / WHAT YOU GET

Built for people who
actually use the plan.

Everything here ships in Midas 0.48.4.
Nothing on this page is a roadmap.

01

Spend at API rates. One total.

Midas prices tokens across your enabled providers at public API rates and adds them into one figure. It is an estimate of the value you got from your plan, not your invoice, and the label says so everywhere it appears.

API-RATE ESTIMATE ≠ YOUR BILL
02

Pick the period that matches your plan.

This month, the last 30 days, or any recorded month, from one dropdown that previews each date range. Arrow keys move, Return selects, Escape closes. Quota reset schedules never change with the period.

NEW IN 0.35.1
03

Every Codex account, side by side.

Sign in to more than one OpenAI account. Each gets its own identity, remaining-usage bar, reset time, and banked reset count. With Track all connected accounts enabled, each account’s reported cloud history appears separately, too. Percentages are never averaged into one misleading number.

NEW IN 0.35.0 · BANKED RESETS IN 0.35.2
04

Account-wide Codex usage, priced automatically.

Turn on Track all connected accounts and Midas combines available Codex cloud token history across your accounts and computers, then prices it with a blended rate from your recent local history. Optional iCloud sharing keeps that rate consistent between Macs. Combined Sync history uses recorded local events instead, so cloud totals are never added twice.

CUSTOM RATE OR TOKENS-ONLY IF YOU PREFER
05

Banked resets, counted per account.

Codex can grant on-demand rate-limit resets. Midas shows how many each account has available, with reported expiry times, separate from the scheduled weekly reset. Zero, unavailable, and last-known are three different states, and they look different.

NEW IN 0.35.2
06

Local first. Native. Signed.

A real SwiftUI app with a menu-bar panel and a resizable Usage window. It reuses the sign-ins you already have, stores no passwords, bundles its provider logos, and lets you turn Keychain access off entirely. Updates are signed, notarized, and delivered through Midas’s own feed.

SWIFT + SWIFTUI / macOS 14+ / APPLE SILICON
07

Claude’s dollars, plus its 5-hour and weekly limits.

Midas values your Claude Code usage at Anthropic API rates, including cache writes and reads, fast mode, and web searches, and adds it to your total. Claude’s 5-hour and weekly limits sit side by side, each with its own bar, % left, and reset time, while a Fable-only weekly limit stays in account details whenever Claude reports one. Sign in to personal and work Claude accounts separately; each gets its own identity, limits, and reset times. Claude Code spend remains a provider-wide estimate from recorded activity, combined across Macs when Sync is enabled.

MULTIPLE CLAUDE ACCOUNTS NEW IN 0.41.2
08

Anthropic API spend, billed and separate.

Add an organization Admin API key to see what your Anthropic API usage actually billed, with token usage by model. It is labeled Billed API spend and kept out of the estimated total, so nothing is counted twice.

NEW IN 0.36.0 / ADMIN API KEY
09

Cursor’s three meters, at a glance.

Cursor Models, Other Models, and Grok Bot each get a bar with % left and reset time, straight from Cursor’s usage dashboard. A Grok Bot trial shows when it ends. Cursor’s spend estimate appears as soon as Midas opens and stays, labeled last known, if cursor.com has a bad moment.

NEW IN 0.40.0
10

Compare your spend, month by month.

Open Spend history to compare each provider’s cost or tokens across 6 or 12 calendar months. Hover for details or read the monthly records. This month compares the same elapsed days in the previous month when matching records are available. Missing history stays unavailable; Codex cloud currently supplies 30 days and Mistral only its current month.

NEW IN 0.42.0 / AVAILABLE HISTORY ONLY
11

Your devices, one usage history.

Choose a private Tailscale hub on your Mac mini, or the same private iCloud Drive folder on your Macs. Midas combines recorded Codex, Claude Code, and Muse Code usage and counts each Cursor account once. A lightweight collector can share supported numeric usage from Linux, including DGX Spark. Monthly charts, menu-bar totals, and widgets use the combined history. No Midas account needed.

TAILSCALE HUB NEW IN 0.47.0 / OPT-IN / UTC DAYS
12

DGX Spark. Local tokens, cloud value.

A dedicated NVIDIA-branded card shows recorded local inference tokens for the last 30 days. For supported, identified models with a verified token mix, it estimates what those tokens would cost at a named hosted API rate. Cached input is part of input, counted once. Collection starts when the metrics collector is enabled; older or unknown usage stays unpriced, with partial coverage shown.

NEW IN 0.47.0 / LOCAL INFERENCE / API-EQUIVALENT ESTIMATE
04 / SETUP

Three steps.
No spreadsheet.

Guided setup opens on first launch. Pause any time;
Continue setup remembers where you were, even after a restart.

  1. STEP 01

    Services

    Choose Codex, Claude, Cursor, and/or Meta. You don’t need a Codex account to use Midas.

  2. STEP 02

    Accounts

    Connect the accounts you use. Codex and Claude support several separate sign-ins. Give each Claude account a nickname and check its verified email before saving. Cursor opens its existing browser sign-in. Meta reads Muse Code activity already on this Mac.

  3. STEP 03

    Overview

    Preview estimates for your chosen period, then choose Open Midas. Background refresh keeps going while you get back to work.

Claude’s 5-hour and all-model weekly limits appear separately for each connected account. Model-specific weekly limits stay in account details when Claude reports them. The Anthropic API, Gemini, GitHub Copilot, OpenRouter, Grok, Perplexity, and 40+ other providers from the CodexBar foundation are in Settings → Providers.

05 / HOW THE NUMBERS WORK

Estimates you can
actually reason about.

Midas would rather show you a gap than a comforting wrong number. These rules apply everywhere a figure appears: panel, Usage window, menu bar, and tooltips.

Midas does

  • Price tokens at public API rates and label them as estimates.
  • Leave missing estimates out and explain coverage in Estimate details.
  • Keep weekly quota separate from session limits.
  • Keep currencies separate instead of converting them.
  • Show zero as zero and unavailable as unavailable.

Midas doesn’t

  • Call an estimate your bill.
  • Turn a missing estimate into $0.
  • Average different quotas into one percentage.
  • Invent a quota for a provider that doesn’t report one.
  • Fetch provider logos from a server. They ship inside the app.
06 / OPEN SOURCE

Read the code.
Change the code.

Midas is free and MIT-licensed, built on Peter Steinberger’s CodexBar. Every pricing rule, every provider probe, and every privacy decision is in the repository. If you disagree with one, open an issue or send a pull request.

Browse the source on GitHub
07 / QUESTIONS

Before you download.

Is “estimated spend” the same as my bill?

No. Midas prices your tokens at public API rates to show what that usage would have cost pay-as-you-go. Subscription fees, credits, discounts, and each provider’s billing rules are different. If an estimate is missing, Midas leaves it out and explains the coverage in Estimate details instead of counting it as $0.

How does Midas put a dollar figure on Codex usage from my other Macs?

With Track all connected accounts on, Midas reads available Codex cloud token history for each connected account and combines it across computers. It prices those tokens with a blended rate taken from your recent priced local history. If it has no usable sample, the dollar figure stays unavailable. You can also set your own rate or show tokens only. Cloud history comes from private ChatGPT analytics endpoints, not an official billing API, so it can change independently of Midas.

Can I see the same combined usage on my Mac mini and MacBook?

Yes. In Settings → Sync, choose a private Tailscale hub on your Mac mini, or the same private iCloud Drive folder on Macs using the same Apple Account. No Midas account is required. Midas combines recorded Codex, Claude Code, and Meta’s Muse Code usage; the same Cursor account is counted once. The hub can also receive supported numeric usage from a Linux collector, including DGX Spark. Shared history uses UTC days and retains up to 365 days of available records. Devices save locally while offline and retry later. Keep the hub awake and reachable for exchanges; saving on one Mac does not confirm the other has received it. Check each device’s last sync. Provider sign-ins and quota limits remain local.

How is DGX Spark’s local inference valued?

The DGX Spark card shows a fixed last-30-days total from newly collected vLLM metrics, with input and output tokens added once. Supported NVIDIA Qwen3.8-Flash-Next NVFP4 usage is compared with Lyceum’s hosted Qwen3.8-Flash-Next rates: $0.20 per million uncached input tokens, $0.05 cached input, and $0.50 output, checked October 17, 2026. The local quantized model and hosted service can differ. This is an API-equivalent estimate, not a bill or a claim of actual savings. Unknown models, missing token mixes, and history from before collection stay unpriced; Midas marks incomplete coverage as partial. DGX Spark has no cloud subscription quota.

Can Midas track Meta AI or the Muse app?

Meta AI and the Muse consumer app are currently unavailable as numeric usage sources. Midas can read and sync Muse Code token logs. It does not estimate tokens from your private conversations or treat missing consumer usage as zero.

Which Macs can run Midas?

Apple Silicon Macs running macOS 14 Sonoma or later. Download the ZIP, unzip it, and move Midas to Applications. Since 0.37.0 Midas is its own app, so it can sit next to an upstream CodexBar install, and your settings from earlier versions come along on first launch.

Do all providers show remaining quota?

Only when the provider reports one. Codex shows weekly remaining, Cursor shows its Cursor Models, Other Models, and Grok Bot meters, and Meta stays out of the quota ring. Observed Meta plan windows can appear in the app; a recorded-activity summary is not a reported plan allowance. Quota percentages are never averaged across accounts or providers.

What is a banked reset?

Codex can grant on-demand rate-limit resets. Midas shows how many each connected account has available, separate from the scheduled weekly reset. Zero means none are available, Unavailable means no count was reported, and stale counts are labeled last known. Hover a row or expand account details to see reported expiry times.

What about Claude, Gemini, Copilot, and the rest?

Guided setup covers Codex, Claude, Cursor, and Meta. Connect personal and work Claude accounts with separate sign-ins and verified emails. Each account shows its 5-hour and all-model weekly limits in the overview; model-specific weekly limits, including Fable when reported, stay in account details. Add an Anthropic Admin API key for billed API spend. Every provider from the CodexBar foundation, including Gemini, GitHub Copilot, OpenRouter, Grok, Perplexity, and more than 40 others, is available in Settings → Providers.

What leaves my Mac?

Midas talks to providers with the credentials you already have and never stores passwords. API keys you add yourself, such as an Anthropic Admin key, stay in your Mac’s Keychain. Provider logos are bundled, so opening the app makes no logo requests. Optional Codex pricing-sample sharing sends derived rates and expires them after 24 hours. Separately, opt-in Sync shares numeric usage history, costs, public model names, opaque account and event IDs, and device labels through your chosen private folder or Tailscale hub. Each hub collector uses a separate revocable usage credential; provider credentials are never shared. Prompts, conversation text, raw emails, and credentials stay local. Anyone with access to that folder can read its usage data; use a private hub and approved devices only.

How do updates work?

Choose Check for Updates in the panel, then Download and Install & Restart. Releases are signed and notarized and delivered through Midas’s own Sparkle feed, which accepts only signed Midas builds. Automatic checks live in Settings → About. Coming from 0.36.0 or earlier? Midas moved to its own app identity in 0.37.0, so download the latest release once and replace your copy; updates are automatic again afterwards, and widgets need to be re-added once.

08 / RELEASE NOTES

What shipped,
release by release.

Every release is signed and notarized, with full notes on GitHub. Newest first.

  1. 0.48.4OCT 20, 2026LATESTIMPROVED

    Every Mac’s full history, shared

    • A busy Mac whose usage history outgrew one sync publication shared only its newest days, so other Macs’ combined totals missed most of its usage. Older days are now shared as daily per-model totals instead of being dropped.
    • Tokens, requests, dollar estimates, and coverage counts for those days are unchanged; only the level of detail shared for older days is reduced.
    • Update each synced Mac so it shares its full history. The private hub and earlier Midas versions read the new totals without changes.
    Release notes
  2. 0.48.3OCT 18, 2026IMPROVED

    Keep Codex on one account

    • An optional local preference limits Codex usage to one OpenAI account. Quotas, cached readings, and cloud history stay with the selected account, without falling back to another login.
    • While scoped, Codex totals use account-verified cloud history and exclude combined device history and local logs that cannot reliably identify the account. History and dollar estimates may be unavailable.
    • Other providers, saved connections, and shared-history contributions keep their existing behavior.
    Release notes
  3. 0.48.2OCT 18, 2026IMPROVED

    AI Spend

    • The overview total now has the shorter heading AI Spend. Help text and More → Estimate details retain its meaning as estimated value at API rates.
    • This is a label change. Amounts, recorded tokens, coverage, and separate billed-spend accounting keep the same rules.
    Release notes
  4. 0.48.1OCT 17, 2026IMPROVED

    A cleaner overview, with less noise.

    • Remove inline Partial total and Partial estimate notices from the overview. Estimate details keeps the full source coverage, missing-history, and pricing explanation one click away.
    • Meta’s recorded-activity fallback uses a compact summary in the overview, with its activity breakdown retained in provider details. Observed plan limits keep their own percentages and resets.
    • Dollar estimates, recorded tokens, separate account limits, and billed-spend exclusions keep the same accounting. Missing data stays unavailable; the interface change does not make an estimate complete or turn it into a bill.
    Release notes
  5. 0.48.0OCT 17, 2026DESIGN

    A compact, native Mac interface.

    • A narrower, content-sized panel keeps the total and actions fixed while provider rows scroll. Each provider keeps its dollar estimate, independent limits, % left, and reset times visible.
    • SF Pro, SF Symbols, system colors, and your Mac’s accent make the interface feel at home. One Liquid Glass navigation capsule appears on macOS 26 and later, with a solid fallback for Reduce Transparency and older systems.
    • Usage and Settings get native sidebars. A native date menu selects the value period, while coverage and sync details open inside the panel with a Back button.
    • Recorded tokens, separate account limits, DGX Spark pricing coverage, and private sync retain their existing rules. Missing data stays unavailable, partial estimates stay labeled, and billed API spend stays separate. macOS 14 remains supported.
    Release notes
  6. 0.47.1OCT 17, 2026IMPROVED

    Lighter shared history and background sync.

    • Shared-history merging and validation use fewer temporary objects and stop re-encoding identical records just to compare them.
    • Local session scans and recurring sync merges and saves run away from the UI thread, with temporary parsing and encoding objects released after each operation.
    • Unchanged Codex cache refreshes skip disk rewrites. New activity, explicit refreshes, and pricing or history-window changes still update usage.
    Release notes
  7. 0.47.0OCT 17, 2026NEW

    DGX Spark joins your shared usage.

    • A dedicated NVIDIA-branded DGX Spark card shows recorded local inference tokens for the last 30 days, with a named hosted-model API-equivalent estimate when the model and token mix are known.
    • A private Tailscale hub combines supported numeric records from Macs and Linux collectors, with device credentials, offline retries, duplicate detection, and per-device sync status.
    • Connected OpenAI accounts keep independent identities, quota rows, and reported cloud history. Combined local event totals do not add that cloud history a second time.
    • Cached input is counted once. Collection is prospective; unknown models and earlier unpriced records remain visible as partial coverage.
    Release notes
  8. 0.46.2OCT 15, 2026NEW

    Much lighter memory.

    • The Muse session-log scan now streams instead of loading whole log files, and Claude/Codex cost scans reuse a shared retained window instead of reparsing every log when the window slides or widens.
    • On a machine with gigabytes of local session logs, the startup memory peak drops from ~3.6 GB to ~129 MB, steady footprint sits near 50 MB, and repeat scans drop from ~26 seconds to under a second. Usage totals are unchanged.
    • Includes 0.46.1: sync stays connected through temporary iCloud identity outages and resumes automatically.
    Release notes
  9. 0.46.1OCT 15, 2026

    Sync stays connected.

    • Sync no longer disconnects when macOS announces an unchanged iCloud account; the folder and offline history survive temporary identity outages and resume automatically.
    • Local usage scans stop reparsing every session log on window slides, and offscreen dashboard WebViews are capped.
    Release notes
  10. 0.46.0OCT 13, 2026

    Meta shows its real plan usage.

    • Current usage and Weekly limit now show Meta’s own % left with each window’s reset time, matching the Meta Model API usage page.
    • Midas reads the windows through the Muse host, checking with one minimal prompt at most once an hour; the recorded-activity estimate remains only as a fallback.
    Release notes
  11. 0.45.0OCT 12, 2026

    Meta bars now show usage left.

    • Today and Last 7 days now read “% left” with tokens used, tokens left, and request counts, draining as Muse usage grows and warning when headroom runs low.
    • Meta reports no plan limit, so “left” is measured against your busiest comparable recorded day/week and every bar names that baseline.
    • Menu-bar rings still mean remaining capacity only for providers that report one; Meta keeps its neutral ring.
    Release notes
  12. 0.44.0OCT 12, 2026

    Meta, with bars like Claude.

    • Meta now shows Today and Last 7 days of recorded Muse tokens in the overview, provider details, and legacy menu card, each with a bar and request count.
    • Each bar is scaled to your busiest comparable recorded window and labeled as a share of it, like “43% of busiest day”.
    • Meta reports no quota, limit, or reset, so Midas never shows it as percent left. Menu-bar rings still mean remaining capacity only for providers that report one.
    Release notes
  13. 0.43.0OCT 12, 2026NEW

    Your Macs, one recorded history.

    • Opt in to Settings → Sync and choose the same private iCloud Drive folder on each Mac. No Midas account is required.
    • Combine recorded Codex, Claude Code, and Muse Code usage while counting each Cursor account once. Monthly charts, menu-bar totals, and widgets follow the selected combined scope.
    • Shared history uses UTC days and keeps available records through offline periods. Credentials and conversation text stay local; cloud delivery can take time.
    • Meta AI and the Muse consumer app remain unavailable as numeric usage sources. Missing data is never counted as zero.
    Release notes
  14. 0.42.0OCT 12, 2026NEW

    See how your months compare.

    • Spend history now charts each provider’s monthly cost or tokens across 6 or 12 calendar months, with hover details and accessible monthly records.
    • Current-month comparisons use the same elapsed days in the previous month when matching records are available. Missing history stays unavailable, and explicit zero stays zero.
    • Load up to 12 months from supported sources. Available history still depends on the provider: Codex cloud currently supplies 30 days; Mistral only its current month.
    Release notes
  15. 0.41.2OCT 4, 2026NEW

    Your Claude accounts, side by side.

    • Sign in to personal and work Claude accounts separately, check each verified email, and add nicknames.
    • Each saved account refreshes independently with its own 5-hour and all-model weekly limits, reset times, and errors. Switching accounts leaves your Claude Code login alone.
    • Model-specific weekly limits stay in account details. Claude Code spend remains a provider-wide estimate from activity on this Mac.
    Release notes
  16. 0.41.1SEP 29, 2026DESIGN

    One ring for the whole week.

    • Orbit’s single outer ring shows your favorite provider’s weekly allowance remaining, or Cursor’s monthly allowance.
    • Fable and other model-specific limits no longer drive the Orbit ring or its low-quota warning. Those limits remain in the provider panel.
    • A missing weekly allowance stays unknown instead of falling back to a session or model quota.
    Release notes
  17. 0.41.0SEP 27, 2026NEW

    Grok Bot stays put, with its own mark.

    • Cursor’s Grok Bot weekly bar now carries a small Grok mark. If cursor.com skips a refresh, Midas keeps the last known reading until the weekly reset, or says Unavailable, instead of dropping the bar.
    • Cursor spend stays with the account that produced it, and switching Cursor accounts refreshes it right away.
    • This month and Last 30 days follow your Mac’s calendar, so the month no longer flips at 5 PM on the US west coast.
    • Codex sign-in is sturdier: overlapping refreshes no longer race on the login token, and each refresh makes one usage request instead of two.
    Release notes
  18. 0.40.0SEP 24, 2026NEW

    Cursor gets three bars and a steadier estimate.

    • Cursor Models, Other Models, and Grok Bot weekly now sit in the overview, each with % left and reset time, like Cursor’s own usage dashboard. No more clicking into Cursor to check Grok Bot.
    • A Grok Bot trial shows when it ends instead of a weekly reset, and plans without Grok Bot no longer show an empty bar.
    • Cursor’s spend estimate shows as soon as Midas opens and stays, labeled last known, when a cursor.com refresh fails.
    • Reads that change while you’re using Cursor are retried, and more Cursor usage gets priced.
    Release notes
  19. 0.39.0SEP 24, 2026NEW

    Claude’s Fable-only limit, at a glance.

    • When Claude gives Fable its own weekly limit, Midas shows it as a third Claude bar, Fable weekly limit, with % left and reset time next to the 5-hour and weekly limits.
    • It matches the Fable only bar in Claude’s own Usage screen, whether Midas reads Claude through its sign-in, claude.ai, or Claude Code.
    • A nearly used-up Fable limit flags the menu bar, like any other limit. Your all-models weekly limit still applies.
    • Claude reset times read more reliably from Claude Code. Spend estimates are unchanged.
    Release notes
  20. 0.38.0SEP 24, 2026NEW

    See what your Claude usage is worth.

    • Claude Code activity on your Mac now shows as estimated spend at Anthropic API rates and counts toward your total.
    • Current Claude models are priced out of the box, including Opus 5.5, Opus 5, Sonnet 5, and Fable 5.1, with their cache discounts.
    • Fast mode, US-only inference, and web searches are priced too. Days with a model Midas can’t price are marked partial, never shown as $0.
    • An estimate of value at API rates, not a bill. Codex, Cursor, and Meta estimates are unchanged.
    Release notes
  21. 0.37.1SEP 23, 2026IMPROVED

    Cursor is back.

    • Cursor spend and quota show again after the move to Midas’s own app identity. Midas now brings over the sign-in sessions saved by earlier versions.
    • macOS may ask once to allow Midas to use the saved session. Declining never brings the prompt back.
    • Coming from 0.36.0 or earlier? Download this release once and replace your copy. In-app updates are automatic afterwards.
    Release notes
  22. 0.37.0SEP 23, 2026IMPROVED

    Midas is now its own app.

    • Midas ships as Midas.app with its own identity, so it no longer shares settings, Keychain items, or storage with CodexBar, and both can be installed side by side.
    • Your preferences and history come along on first launch. macOS may ask once per provider to reuse saved sign-ins.
    • Coming from 0.36.0 or earlier? Download this release once and replace your copy. In-app updates are automatic again afterwards.
    • Re-add your Midas widgets once after installing. Quota, spend, and reset logic are unchanged.
    Release notes
  23. 0.36.0SEP 22, 2026NEW

    Claude limits at a glance, Anthropic API spend, and a Settings fix.

    • See Claude’s 5-hour limit and weekly limit together, each with its own bar, % left, and reset time.
    • New Anthropic provider: add an organization Admin API key to see billed API spend and token usage by model, next to your Claude subscription limits.
    • Billed spend is labeled Billed API spend and kept out of the estimated total, so nothing is counted twice.
    • Settings opens reliably again on macOS 27, even after Midas has been running for days.
    Release notes
  24. 0.35.4SEP 18, 2026IMPROVED

    Golden Gate-ready updater and hotkeys.

    • Update Sparkle to 2.10.0, the Golden Gate compatibility release, for signed in-app updates on macOS 27.
    • Update KeyboardShortcuts to 3.1.0, fixing the release-build crash with the Swift 6.3 compiler.
    • Clear the format/lint gate. No behavior changes: quota, spend, and reset logic are unchanged.
    Release notes
  25. 0.35.3SEP 15, 2026IMPROVED

    Inline period details and last-known Codex quota.

    • Choose This month or Last 30 days from an inline spend-period list, so the panel no longer crashes on macOS 27.
    • Open estimate-coverage details, including Partial total, inline instead of a nested popover.
    • Show last-known Codex weekly remaining and banked resets immediately for managed accounts that omit workspaceAccountID.
    • Development and Apple Development builds can download the signed GitHub ZIP from Check for Updates. Sparkle Install & Restart still needs a Developer ID copy.
    Release notes
  26. 0.35.2SEP 12, 2026NEW

    Banked resets for every account.

    • See how many on-demand rate-limit resets each connected Codex account has, in the overview and in account details.
    • Zero and Unavailable are shown differently, and stale counts are labeled last known with reported expiry times.
    • Reset counts stay visible even when weekly quota data is missing.
    • The menu-bar panel re-anchors every time it opens, so it no longer lands in a stale position.
    Release notes
  27. 0.35.1SEP 12, 2026IMPROVED

    A better spend-period picker.

    • Choose This month, Last 30 days, or any recorded month from a dropdown anchored to its button, in both the overview and the Usage window.
    • Every option previews its date range. Use the arrow keys to move, Return to select, and Escape to close.
    • Recorded months come from your displayed providers’ history. Changing the period changes spend estimates, never quota reset schedules.
    Release notes
  28. 0.35.0SEP 12, 2026NEW

    Multiple Codex accounts, cloud activity, and guided setup.

    • Sign in to several Codex accounts, each with its own remaining-usage bar and reset time. Local spend history stays put when you switch accounts.
    • Combine available Codex cloud token history across connected accounts and computers, and price it automatically from your recent local history. Optional iCloud sharing, custom rates, and a tokens-only mode are included.
    • New three-step setup: Services, Accounts, Overview. Pause and resume any time.
    • The overview now leads with estimated dollars for a calendar period or the last 30 days. Missing estimates stay unavailable and partial totals say so.
    • Fixed Codex account rows that wrongly showed unavailable usage, and kept settings responsive by avoiding Keychain reads while rendering.
    Release notes
  29. 0.33.6SEP 5, 2026DESIGN

    A brighter crown.

    • The pixel crown’s highlight sweep is brighter and leaves a warm trail.
    • One crown tooth sparkles after each sweep, with a longer rest in between.
    • Reduce Motion keeps the crown completely still.
    Release notes
  30. 0.33.5SEP 5, 2026IMPROVED

    A tighter menu-bar item.

    • The status item now sizes to its visible content instead of reserving fixed width.
    • Less empty space beside your spend and quota readings, with digits that keep their width as values change.
    Release notes
  31. 0.33.4SEP 5, 2026DESIGN

    Native typography and the pixel crown.

    • System typography with monospaced digits for money and quota readings.
    • A gold pixel crown with a subtle looping shimmer beside the Midas wordmark.
    • Favorite provider and Check for Updates moved into a fixed panel footer.
    Release notes
  32. 0.33.3SEP 5, 2026NEW

    Signed in-app updates and Orbit.

    • Check for Updates with a standard Download and Install & Restart flow, on a Midas-only signed update feed.
    • Orbit: total spend followed by one favorite provider’s logo and remaining-capacity ring in the menu bar.
    • Switch the favorite provider from the panel dropdown.
    Release notes
  33. 0.33.2SEP 5, 2026IMPROVED

    Honest labels.

    • The primary spend figure is labeled “Token spend (API rates)” in every card, total, tooltip, and accessibility description.
    • Estimates are always presented as estimates, never as your bill.
    Release notes
  34. 0.33.1SEP 5, 2026NEW

    The first Midas menu bar.

    • Ledger, Focus, and Constellation display modes, plus menu-bar spend privacy.
    • Midas Air: a native overview panel and a resizable Usage window with per-provider spend totals.
    • Codex weekly remaining front and center, kept separate from session limits.
    Release notes
Full release history on GitHub
FREE. OPEN SOURCE. READY NOW.

Stop guessing
what you’ve used.

Download Midas 0.48.4

APPLE SILICON / macOS 14+ / SIGNED & NOTARIZED / MIT

  1. 1. Download the ZIP
  2. 2. Move Midas to Applications
  3. 3. Follow the three-step setup