← all repositories
mm7894215/TokenTracker

An Itemized Bill for Your AI Coding Assistants

TokenTracker exists to end the guessing game of tracking token spend across two dozen AI coding tools, entirely on your own machine.

1.1k stars JavaScript LLMOps · EvalCoding Assistants
TokenTracker
Velocity · 7d
+5.9
★ / day
Trend
cooling
star history

What it does TokenTracker is a local-first usage monitor that hooks into 22 AI coding assistants — from Claude Code to Cursor to Grok Build — to collect token counts and estimate costs. It aggregates everything into a browser dashboard and optional native desktop widgets, running entirely offline without accounts or API keys. A built-in cost engine tracks 2,200+ model prices via LiteLLM with daily updates and offline caching, though it notes when vendors haven’t published rates.

The interesting bit The integration strategy is admirably pragmatic: it auto-writes SessionEnd hooks into tool configs where possible, ships its own plugins for extensible tools like OpenCode, and falls back to passively reading SQLite databases and JSONL logs the tools already produce. It is essentially sophisticated glue code — but the kind that saves you from manually tallying usage across two dozen disparate CLIs.

Key highlights

  • Supports 22 tools via three integration patterns: config hooks, bundled plugins, and passive log readers
  • Runs entirely local; collects only token counts and timestamps, never prompts or file contents
  • Cost engine covers 2,200+ models with daily price updates and a bundled offline snapshot for air-gapped use
  • Optional opt-in leaderboard and a Skills tab for browsing and syncing 250+ public agent skills across compatible tools
  • Native macOS menu bar app with desktop widgets, plus a Windows system tray version

Caveats

  • Grok Build token splits (prompt vs. output vs. cache) are currently estimated from limited local telemetry, not exact per-call data
  • Some niche models like Tencent’s hy3-preview show $0 cost because the vendor hasn’t published pricing
  • Desktop app is native only for macOS and Windows; Linux users stick to the CLI and browser dashboard

Verdict Worth a look if you juggle multiple AI coding tools and want to audit spend without cloud dependencies. Skip it if you only use one tool with its own built-in usage page and don’t care about cross-tool totals.

Frequently asked

What is mm7894215/TokenTracker?
TokenTracker exists to end the guessing game of tracking token spend across two dozen AI coding tools, entirely on your own machine.
Is TokenTracker open source?
Yes — mm7894215/TokenTracker is open source, released under the MIT license.
What language is TokenTracker written in?
mm7894215/TokenTracker is primarily written in JavaScript.
How popular is TokenTracker?
mm7894215/TokenTracker has 1.1k stars on GitHub and is currently cooling off.
Where can I find TokenTracker?
mm7894215/TokenTracker is on GitHub at https://github.com/mm7894215/TokenTracker.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.