GH Copilot Usage Tracker

Track your GitHub Copilot usage by model: AI credits, input/output tokens, and monthly usage totals.

GH Copilot Usage Tracker works with GitHub Copilot Chat in VS Code, Copilot agent sessions, the Copilot SDK, and Copilot CLI using usage records available on your local device. It adds a compact Copilot credits indicator to the bottom-right status bar, with a small hover summary and click popup—not a Terminal or docked panel.

This independent community extension is not an official GitHub product or an account-wide billing report. Estimates and partial data are labeled. Nothing is uploaded to a usage-collection service.

The status bar is used because stable VS Code APIs support rich native hover content there, but do not expose a custom hover window attached to the chat-toolbar contribution. No private APIs or renderer injection are used.

The extension reads one configured provider at a time:

• GitHub Copilot — supported default: VS Code chat, agent sessions, SDK and CLI.
• Claude Code — preview provider: local Claude Code transcripts.

Use GH Copilot Usage Tracker: Select Chat Provider from the Command Palette or the aiUsageCreditsMeter.provider setting. This release defaults to GitHub Copilot and does not mix providers in one total. A Claude or Gemini model used inside GitHub Copilot still counts as Copilot usage; the separate Claude Code provider is an explicit preview only.

Install

Install from the Visual Studio Marketplace, or download the .vsix from the project releases, then run Extensions: Install from VSIX... and reload the window. Requires desktop VS Code 1.141 or newer. This is an independent community extension, not an official GitHub or Anthropic product.

The extension ID remains apkola29.ai-usage-credits-meter so existing installs update normally; its visible name is GH Copilot Usage Tracker. Command and setting IDs are unchanged for compatibility.

What you see

Hover: a small native tooltip shows the cycle total and the five models with the highest credit usage. It disappears when the pointer moves away.

Click: a compact floating list shows the five highest-used models for the configured ranking. Close it using ×, Esc, or clicking outside. Use Show all models only if you want the longer list.

• Each compact row shows credits and tokens. Selecting a model copies its row.
• The popup has Refresh and a tokens/credits ranking toggle.
• Command Palette: GH Copilot Usage Tracker: Show Compact Usage Popup opens the same small list.
• Show Detailed Dashboard remains optional in the Command Palette; it is never opened by a normal hover or click.
• Use My GitHub Account: reuse VS Code's GitHub sign-in rather than entering a PAT or username.

≈ marks numbers that include an estimate (see Accuracy).

A single summary scan warms the hover after startup. Further scans are on demand; there is no polling loop. Cached summaries expire after one minute and are invalidated when the provider, settings, or GitHub sign-in changes.

Where the numbers come from

Usage records are read locally and read-only. The tracker does not upload prompts, transcripts, usage records, or credentials. Optional account connection uses VS Code's native GitHub authentication broker; it is not an independent password/PAT login.

Identity and safety

• Reuses the existing VS Code GitHub identity with only read:user permissions. If several accounts are present, select the account you use with Copilot; the tracker does not guess.
• Authentication tokens are never sent to the worker, tooltip, popup, or optional webview, persisted by the tracker, or written to its logs. Only the chosen account ID preference and displayed account label are used.
• Historical local records are device-scoped, not guaranteed account-scoped. Untagged records cannot prove which GitHub account incurred them; this is stated in the UI. These numbers are not an authoritative account-wide invoice.
• Reads the current VS Code profile and the current OS user's Copilot stores. Provider/data-directory settings are machine-scoped, not controlled by a repository.
• Disabled in untrusted and virtual workspaces. No production HTTP listener, debugging port, child shell, or third-party collector.
• Native hover Markdown is untrusted and has HTML disabled; local model/account text is escaped. The optional dashboard still has scripts disabled, no local resource access, nonce-protected styles, and a CSP blocking scripts and network connections.
• One lazy worker, at most four queued scans, duplicate-request coalescing, a 120-second timeout, and termination after 30 seconds idle or extension deactivation.
• Worker heap limits are 512 MiB old-generation and 32 MiB young-generation. The parsed-summary cache is limited to 256 files; prompts/transcript bodies are not retained in that cache.
• Read/parse failures are visible. Future changes to VS Code or Copilot's local formats still require regression testing; no implementation can guarantee that future releases will never introduce an issue.

Billing basis

• Since 2026-06-01, Copilot bills in GitHub AI Credits, priced per token per model.
• For the Claude models checked (Opus 4.7–5, Opus 5.5, Sonnet 5), GitHub's per-token billing equals Anthropic's list price, so the tools are directly comparable.
• The billing cycle is the calendar month in UTC; set aiUsageCreditsMeter.cycleResetDay if yours differs.

Accuracy

• Copilot. Finished agent/CLI runs and VS Code chat requests are exact for credits.
  • VS Code chat input tokens are the final prompt size of each request, so they are a lower bound.
  • Long agent sessions that are still running, or that ended without a clean shutdown, only expose a session-level credit total. That total is split across models using each call's model tag and marked ≈.
• Claude Code.
  • Each message is counted once: streaming duplicates and messages copied into resumed sessions are removed.
  • Credits reflect API list price, not your Claude subscription.
  • Claude Code deletes transcripts after 30 days by default (cleanupPeriodDays).
• Scope. Only usage on this machine is included.

Settings

Development

npm test               # fixture tests (node:test)
npm run usage          # print the ranking in a terminal (add -- --validate to cross-check Copilot sources)
npm run test:vscode    # integration test in an isolated VS Code instance (add -- --vsix <file> to test a package)
npm run test:vscode -- --agents # run it in the Agents/Sessions window
npm run test:popup     # real hover -> mouse-leave dismissal; click -> close/Escape/click-away
npm run package        # build the .vsix offline
npm run package:vsce   # standard vsce package and its credential/file checks
node test\toolbar.js --vsix ai-usage-credits-meter-0.4.2.vsix # same native-UI test on the installed package
code --install-extension ai-usage-credits-meter-0.4.2.vsix

Requires VS Code 1.141+. It uses the built-in node:sqlite, so there are no native dependencies. UI-click tests launch an isolated profile with a temporary loopback debugging port and stop only their owned processes (matching PID and creation time) on exit; those test scripts are excluded from the VSIX.

Keep older installed-version folders until VS Code has reloaded. Let VS Code's extension manager clean them up; manually deleting a version still cached by an open window breaks activation.

Public source and privacy

Source project and issue tracker. See the repository privacy policy and security policy. Only synthetic fixtures are included in public tests; no personal usage data or session logs are bundled.

Licensed under MIT.