native macOS · open source · local-first

Quota & cost,
quietly in your menu bar.

Vibe Bar is a native macOS menu-bar app for people who run coding agents all day. Quota forecasts, local token cost, and service status for ChatGPT, Claude, Gemini and Grok — plus usage cards for twenty more providers. No dashboard required.

macOS · Apple Silicon · ~6 MB

macOS 26+Swift 6.2AGPL-3.0~50 MB RAM
/ INTERACTIVE PRODUCT UI

The app itself, rebuilt in the page.

Click through the same surfaces Vibe Bar uses on macOS: the tabbed popover with all six pages, the floating mini window, and settings. The numbers are staged, but the interaction model is real.

Vibe Bar
Wed Jul 22 15:19

Overview

All providers · quota & cost · Updated just now

CostUpdated 6 seconds ago
Total cost$50039Total tok52.25BPeak day$2498Peak day tok1.35BToday$323Yesterday$8077-day$485430-day$26289Today tok373.41MYesterday tok991.18M7-day tok6.17B30-day tok26.99B
All Providers Cost History
$4000$2000$0.00
Jun 23Jun 28Jul 3Jul 8Jul 13Jul 18
Grok Cost25 sec
Today$4.105.88M tokYesterday$6.429.14M tok7D$21.1530.42M tok30D$69.3798.75M tokAll$73.10104.20M tok
Top model 7Dgrok-4-code$18.90 · 27.1M tok · 89% share
Cost History
$10$5$0.00
Jun 23Jun 28Jul 3Jul 8Jul 13Jul 18
Gemini Cost25 sec
Today$1.73892.6k tokYesterday$2.271.75M tok7D$15.1010.33M tok30D$174104.14M tokAll$207196.09M tok
Top model 7Dgemini-default$14.95 · 9.99M tok · 99% share
Cost History
$40$20$0.00
Jun 23Jun 28Jul 3Jul 8Jul 13Jul 18
All providers — Past Year$50039 total
JulAugSepOctNovDecJanFebMarAprMayJunJul
MonWedFri
LessMore
StatusUpdated just now
OpenAIDegradedElevated Errors Affecting File Uploads99.97%
AnthropicDegradedElevated errors for Claude Code98.82%
GoogleUpAll services operational93.33%
xAIUpAll services operational98.10%
ChatGPT Cost25 sec
Today$153206.16M tokYesterday$311463.66M tok7D$19783.15B tok30D$1493512.89B tokAll$2938026.85B tok
Top model 7Dgpt-5.6-sol$1880 · 2.54B tok · 95% share
Cost History
$2000$1000$0.00
Jun 23Jun 28Jul 3Jul 8Jul 13Jul 18
Claude Cost25 sec
Today$168166.05M tokYesterday$492525.02M tok7D$28423.00B tok30D$1111013.93B tokAll$2037925.15B tok
Top model 7Dclaude-fable-5$1705 · 1.25B tok · 60% share
Cost History
$2000$1000$0.00
Jun 23Jun 28Jul 3Jul 8Jul 13Jul 18
Model RankingAll providers · all time
#1
gpt-5.5
$886629%
#2
claude-opus-4-8
$576719%
#3
claude-fable-5
$519517%
#4
gpt-5.6-sol
$374612%
#5
claude-opus-4-7
$28129%
#6
claude-sonnet-5
$14455%
When you use everythingPeak 6pm · Thu 12am
3.85B1.92B0
12am6am12pm6pm
Sun
Mon
Tue
Wed
Thu
Fri
Sat
QuietHeavy
/ STRAIGHT FROM THE APP

Real screenshots, zero staging.

Captured from the running app — the same popover, mini window and settings you get after installing.

Overviewscroll ↓
Overview — all providers at a glance
/ FEATURES

Every number that matters,
none of the noise.

Live local data — quota forecasts, cost, model rankings, heatmaps and service status — pulled straight from your CLI session logs and provider quotas.

01 · FORECAST

Forecasts, not just percentages.

Every window gets a personal forecast — time-only pace vs. projected-at-reset, with a confidence label — so you know whether you'll run out before the reset, days ahead.

chatgpt
91%
claude
65%
gemini
89%
grok
11%
02 · COST

Local-first cost.

Computed from your own session logs — never sent anywhere.

today
$6.22
+$0.84
7 days
$42.30
30 days
$186
all time
$1,402
03 · HEATMAP

A year of work, in one row.

Contribution-style heatmap of your actual coding-agent activity — spot the streaks, spot the slumps.

52 weeks agolessmoretoday
04 · BURN RATE

Hourly burn for today.

Same-day, hour-by-hour pace. See the spike, then trace it back.

05 · MODELS

Top models, by usage.

Which model ate your tokens this week.

#1claude-opus-4-7
$1499
#2gpt-5-codex
$586
#3claude-haiku-4
$14.50
#4claude-sonnet-4
$4.92
06 · COST HISTORY

Thirty days of spend.

Daily totals with average and peak — see the surge before payday hits.

$80$40$0
avg$28
total$842avg / day$28.07peak$80
07 · WHEN YOU SHIP

Hour-by-hour, week-by-week.

168 cells of when you actually use coding agents — find your real working hours.

00061218
quietheavy
08 · STATUS

Service status, on tap.

Pulled live from all four providers' status pages — per-component 90-day uptime stripes included.

OpenAIAll Systems Operational
99.97%uptime
AnthropicAll Systems Operational
98.82%uptime
GoogleAll services operational
93.33%uptime
xAIAll services operational
98.10%uptime
/ FOUR CORE PROVIDERS + MISC

Four first-class tabs,
plus everything else.

ChatGPT, Claude, Gemini and Grok each get a full detail page — quota forecasts, cost, rankings, heatmaps, service status. Twenty more providers show up as usage-only cards in the Misc tab.

ChatGPT / Codex

~/.codex/sessions/*.jsonl
  • Weekly + Codex Spark windows, reset credits
  • Per-component OpenAI status with 90-day uptime stripes
  • Local Codex session cost & token history

Claude / Claude Code

~/.claude/projects/**/*.jsonl
  • 5-hour, weekly and per-model windows (Sonnet, Opus, Fable…)
  • Cookie minimization — only sessionKey, in Keychain
  • Local Claude Code session cost & token history

Gemini / AntiGravity

Gemini Web + AntiGravity
  • Gemini Web + AntiGravity unified in one tab
  • Separate Gemini vs. Claude+GPT model buckets
  • Combined Gemini cost & usage analytics

Grok / xAI

SuperGrok · Grok Build
  • SuperGrok weekly credits with run-out forecasts
  • xAI service status & uptime
  • Grok Build session cost tracking

And the Misc wall

Usage-only cards — sign in or paste a key. 5-hour / weekly / monthly windows with reserve & deficit pacing.

CopilotCursorKimiZhipu GLMMiniMaxBailianVolcengineTencent HunyuanBaidu QianfanXiaomi MiMoiFlytek SparkOpenCode GoOllamaOpenRouterKiloKiroWarp+ more
/ PRIVACY · LOCAL-FIRST

Your data never leaves your Mac.

Vibe Bar reads your local CLI credentials and session JSONL logs — read-only. It never touches the originals. Derived cost and token history stays under ~/.vibebar/, with retention controls and a one-click clear.

Read-only access to CLI session files
Claude cookies minimized to sessionKey, stored in Keychain
Privacy Mode keeps cost data off disk entirely
Configurable retention, defaults to forever
~/.vibebar/ ├── settings.json # preferences ├── quotas/ # live quota snapshots ├── cost_snapshots/ # derived cost ├── scan_cache/ # parser cache ├── service_status.json # provider status └── cost_history.json # aggregated
/ BUILD FROM SOURCE

Three commands, one menu-bar app.

SwiftPM-native. The packaging script builds the executable, assembles .build/Vibe Bar.app, and ad-hoc signs the bundle for local use.

01 / TEST

Run the suite.

Parsers, settings, pricing, privacy persistence, usage utils.

swift test
02 / BUILD

Package the bundle.

Default release. Pass debug for a debug build.

./Scripts/build_app.sh
03 / OPEN

Launch it.

App icon copied in, ad-hoc signed, ready to live in the menu bar.

open ".build/Vibe Bar.app"
/ FAQ

Questions, answered.

No. Everything runs locally on your Mac. Vibe Bar reads your CLI session logs and credentials, computes cost and quota state in-process, and writes derived data only under ~/.vibebar/. There's no account, no telemetry, no cloud.
Vibe Bar doesn't ask for keys inside the app. It reads your existing local CLI credentials directly — the same ones Codex CLI and Claude Code already use — so once your CLI works, Vibe Bar works.
Four core providers get full detail pages: ChatGPT/Codex, Claude Code, Gemini (Gemini Web + AntiGravity) and Grok. On top of that, the Misc tab tracks usage for twenty more — Copilot, Cursor, Kimi, Zhipu GLM, MiniMax, Bailian, Volcengine, Tencent Hunyuan, OpenCode Go, Ollama, OpenRouter, Warp and friends.
Yes. Settings → Cost Data has a manual "Clear Cost Data" action. Privacy Mode goes further: it wipes local cost history, snapshots, and scan cache, then keeps cost data off disk entirely while enabled.
The prebuilt release zip targets Apple Silicon (arm64). On Intel, build from source with ./Scripts/build_app.sh — Vibe Bar requires macOS 26+ and the Swift 6.2 toolchain either way; the bundle is ad-hoc signed for local use.
AGPL-3.0 only. You can read it, fork it, and run it. If you host a modified version as a service, the AGPL terms apply.

Make your menu bar useful again.

Open source, local-first, and built for developers who run two agents at once. Star the repo, build it from source, send a PR.