ccbar — Claude Code Bar

A local-first desktop meter for Claude Code CLI and Codex CLI usage. Reads your token usage, cost, and quota status directly from ~/.claude/ and ~/.codex/ on your machine — no cloud, no proxy, no telemetry.

Status: v0 / pre-release. APIs, schemas, and config paths can still shift.

What it does

• Provider tabs — switch between Claude (Anthropic) and Codex (OpenAI) accounts in one window.
• Today / 7-day / 30-day token totals with cost estimates derived from current published list prices.
• Per-model breakdown with cache-hit accounting (input / cache_read / cache_write / output totals separated, since cache reads are cheaper and cache writes are more expensive).
• Quota windows — 5-hour and weekly / 7-day rate-limit status pulled from real headers when available, with a graceful local-token fallback.
• System tray popup — left-click the tray icon for a 340×280 quick view; right-click for Open / Quit.
• Login gate — detects whether each provider's CLI has been used at least once, and surfaces an OAuth flow for Codex when needed.

How it gets data

ccbar reads files on your disk that the official CLIs already write:

• Claude: ~/.claude/usage.jsonl (rate-limit headers when present) and ~/.claude/projects/<encoded-cwd>/<sessionId>.jsonl (assistant messages with detailed usage payloads).
• Codex: ~/.codex/sessions/**/*.jsonl and ~/.codex/archived_sessions/*.jsonl (event_msg.token_count events, including rate_limits.{primary,secondary}).

There is one optional network call: a tiny /v1/messages ping (Claude Haiku, max_tokens=1, ~$0.000007 per call, cached 60s) so Anthropic returns the live anthropic-ratelimit-unified-{5h,7d}-* response headers. Set CCBAR_LIVE_QUOTA=0 to disable it; everything else still works using local data.

ccbar does not:

• Call any usage / billing API,
• Read ~/.claude/quota-status/sessions/* or ~/.codex/auth.json,
• Send your token / data anywhere except the optional Anthropic ping above,
• Persist your credentials. The Claude OAuth token is held in memory only for the lifetime of a single HTTPS request and is never logged or emitted over Tauri events. Codex OAuth tokens (issued by ccbar's own PKCE flow) are stored encrypted-at-rest by the OS via your Tauri app config dir.

Stack

• App shell: Tauri 2 (Rust backend + WebView frontend)
• Frontend: React 19, TypeScript, Vite, Tailwind CSS v4, Recharts, Phosphor Icons, Framer Motion
• Backend: Rust (serde, serde_json, walkdir, chrono, dirs, anyhow, reqwest with rustls-tls, tokio)
• Marketing site (web/): Astro 5 + Tailwind v4

Build from source

Requires:

• Node.js 20+ and pnpm 9+
• Rust 1.80+ (rustup default stable)
• Platform-specific Tauri prerequisites (see tauri.app/start/prerequisites). On Windows you need the Visual C++ Build Tools.

git clone https://github.com/riftzen-bit/ccbar
cd ccbar
pnpm install

# Run the app (frontend dev server + Tauri shell)
pnpm tauri dev

# Or production build:
pnpm tauri build

Other scripts:

pnpm dev          # Vite-only frontend dev (no Tauri shell)
pnpm dev:mock     # Vite + mocked backend (fixture data, useful for design iteration)
pnpm build        # tsc + vite build (frontend type-check + bundle)

cd src-tauri
cargo check       # Rust type-check
cargo test        # Rust test suite

The marketing site lives in web/:

pnpm web:dev      # Astro dev server at http://localhost:4321/
pnpm web:build    # static build
pnpm web:capture  # regenerate app screenshots into web/public/screenshots/
pnpm web:test     # Playwright visual-regression smoke

Privacy

• All usage data stays on your machine.
• No analytics. No telemetry. No remote logging. No silent uploads.
• The opt-in Anthropic ping (above) carries an OAuth token over HTTPS; it is not stored, logged, or passed to the renderer.
• Codex OAuth flow uses the public PKCE client ID app_EMoamEEZ73f0CkXaXp7hrann published by the open-source Codex CLI — there is no separate ccbar identifier presented to OpenAI's auth server.

Limitations

• Frontend tested on Windows; macOS / Linux Tauri builds should work but are not yet smoke-tested per-release.
• OpenAI does not expose a public usage API for ChatGPT Plus / Pro / Codex Pro plans, so Codex live-quota pinging is deferred to a later version. Quota numbers come from rate_limits already embedded in the local JSONL.
• The web/ marketing site form on /signup is a stub — there is no backend yet.

Contributing

Issues and pull requests welcome. Please file an issue before starting on anything large.

License

Apache License 2.0 — see LICENSE for full text. Copyright 2026 riftzen-bit.