Copilot Token Tracker

Track GitHub Copilot Chat token usage directly inside VS Code.

This extension collects usage by calling Copilot Chat's internal export command:

Then it parses successful request metadata (metadata.usage) and keeps model-based cumulative totals in persistent extension storage.

Features

Activity Bar sidebar tab (Copilot Stats) with:
- DASHBOARD (table-based overview: today/month/last month/projected year)
- SESSION SUMMARY
- REQUEST HISTORY (expandable request detail items)
- DAILY SUMMARY / DAILY HISTORY (separate Daily page)
Status bar summary for top model usage
Model-based cumulative totals
Estimated API cost by model (based on public list pricing)
Global cumulative totals across all workspaces (live token data in current VS Code profile)
Recent request list (token breakdown per request)
Persistent state (globalState) across VS Code restarts
Commands for detail view, refresh, and reset

Copilot Token Tracker: Show Usage
- Command ID: github-copilot-token-tracker.showUsage
- Opens Output channel report (model totals, recent requests, totals)
Copilot Token Tracker: Refresh Usage
- Command ID: github-copilot-token-tracker.refreshNow
- Forces an immediate poll/export/parse cycle
Copilot Token Tracker: Reset Usage
- Command ID: github-copilot-token-tracker.resetUsage
- Clears all tracked usage data

GitHub Copilot Chat must be installed and expose the internal export command above.
If the command is unavailable, the status bar shows Unavailable and polling retries automatically.

Data source is an internal Copilot debug/export command and can change between Copilot versions.
Estimated costs use public API list pricing references for OpenAI/Anthropic/Google models (as-of implementation date).
Actual Copilot billing may differ from direct API list pricing.
Deduplication uses request ID priority:
- requestId
- serverRequestId
- ourRequestId
- fallback to log item ID
Usage counts aggregate successful requests only (ChatMLSuccess) and only for request types:
- ChatCompletions
- ChatMessages
- ChatResponses
Live token scope is global per active VS Code profile/user data directory, not per workspace.
True retroactive all-time history before this extension started collecting is not available from this tracker.
If maxProcessedRequestIds is exceeded, very old request IDs are pruned; if those IDs reappear later they can be counted again.
Collector also attempts to read ccreq:latest.request.json as a best-effort supplemental source.
For richer live token capture, increase github.copilot.chat.debug.requestLogger.maxEntries in VS Code settings (Copilot internal request-log ring buffer size).