AI Usage Status Bar

A minimal VS Code extension that shows GitHub Copilot, ChatGPT, Cursor, and Claude usage directly in the status bar — no browser needed.

English | 中文

Changelog | 中文更新日志

Features

• 4 providers — GitHub Copilot quota, ChatGPT/Codex quota windows, Cursor usage, and Claude usage in one place (Claude is temporarily disabled in marketplace builds)
• Reset countdown prefix — each item starts with an Xd countdown to the next reset point
• Codex dynamic windows — status bar uses remaining-to-reset labels (for example 3h and 6d) with remaining %
• Cursor compact usage — status bar shows Auto/API remaining % and appends OD amount when available
• Claude dual-window usage — shows 5h/7d remaining percentages when OAuth usage data is available (auto-adapts to utilization in 0-1 or 0-100 format)
• Clear provider prefixes — uses $(github) for Copilot, $(openai) for ChatGPT/Codex, a stable ◈ fallback for Cursor, and ◆ for Claude
• Unified external format — all providers display remaining quota first in a compact format
• Color warnings — warns on low remaining quota (Copilot / Codex / Cursor)
• Hover tooltip — detailed breakdown on hover for each provider, including detected signed-in account
• ChatGPT update hint — ChatGPT/Codex tooltip includes a "last updated" timestamp
• Auto-refresh — updates every 30 minutes in the background
• Style switching — toggle between minimal (32/50) and verbose (Copilot 32/50) via settings
• Per-provider toggle — enable or disable each provider's status bar item independently

What is tracked

Codex window usage is extracted from local Codex logs (latest API response headers), so it appears after you use Codex at least once.

Installation

From source (development)

1. Clone this repo
2. Open VS Code
3. Press Cmd+Shift+P → Extensions: Install from Location
4. Select the cloned folder

From VSIX (once published)

code --install-extension ai-usage-status-bar-1.0.3.vsix

Requirements

• VS Code 1.74+
• GitHub account signed in to VS Code with Copilot enabled
• OpenAI Codex CLI installed and signed in (for ChatGPT info)
• Cursor installed and signed in (for Cursor info)
• Claude: experimental feature, controlled by one switch.

Platform Support

Cursor's state.vscdb is resolved per platform automatically:

• macOS: ~/Library/Application Support/Cursor/User/globalStorage/state.vscdb
• Windows: %APPDATA%/Cursor/User/globalStorage/state.vscdb
• Linux: ~/.config/Cursor/User/globalStorage/state.vscdb

Codex paths (~/.codex/) and Copilot API calls are cross-platform by default. Claude uses browser OAuth which is cross-platform.

Settings

Search "AI Usage" in VS Code Settings, or edit settings.json directly:

{
  // "minimal" (default): icon + numbers only
  // "verbose": include provider name
  "aiUsage.style": "minimal",

  // Toggle each provider's status bar item
  "aiUsage.providers.copilot": true,
  "aiUsage.providers.chatgpt": true,
  "aiUsage.providers.cursor": true,

  // Optional: Claude OAuth token (legacy, prefer command-based auth)
  "aiUsage.claude.oauthToken": "",

  // Single switch: enable Claude usage experimental feature
  "aiUsage.experimental.enableClaudeUsage": false
}

Style examples:

Settings take effect immediately without reloading.

Commands

How it works

• Copilot: calls vscode.authentication.getSession('github', ['read:user']) → queries api.github.com/copilot_internal/user (undocumented internal endpoint, may change)
• ChatGPT/Codex: reads ~/.codex/auth.json for plan/subscription and ~/.codex/logs_1.sqlite for x-codex-* usage headers
• Cursor: reads state.vscdb (SQLite) for the Bearer token → queries api2.cursor.sh/aiserver.v1.DashboardService/GetCurrentPeriodUsage for Auto/API percentages (falls back to auth/usage when needed); the status bar keeps a stable ◈ prefix for compatibility across VS Code themes/versions
• Claude: controlled by one experimental switch: aiUsage.experimental.enableClaudeUsage (default false). When enabled, data source order is: (1) local Claude session JSONL token-count (5h window), (2) local session rate_limits, then (3) OAuth usage API. OAuth is now a fallback source and includes transient-failure handling (429 cooldown, cached last-success result, and no-data soft fallback). OAuth token resolution is: VS Code SecretStorage first, then Claude Desktop local cache on macOS, then manual setting/env token fallback. OAuth uses PKCE in browser (https://claude.com/cai/oauth/authorize, local callback http://localhost:<port>/callback) and queries api.anthropic.com/api/oauth/usage when local session data is unavailable.

Acknowledgements

• Inspired by the excellent implementation ideas in duddudcns/ai-usage-statusbar, especially around provider fallback strategy and robustness for local usage signals.

License

MIT