You use Claude Code for writing, Codex for running tasks, Cursor for fixing bugs, OpenClaw for automation — then look at your monthly bill and don't know whether to cry or laugh.
junhoyeo/tokscale exists to solve exactly this problem. A Rust-written CLI tool that automatically pulls token consumption data from all your coding agent platforms and piles them together for you to see.
Supported platforms
Currently covers 14:
OpenClaw (Clawdbot/Moltbot), Claude Code, Codex, Cursor, AmpCode, Gemini CLI, Pi.dev, Factory Droid, Kimi, OpenCode — basically all the mainstream coding agents are included.
What it does
The core function is usage tracking: how many tokens each platform spent, which time period burned fastest, which model costs the most. After pulling the data, it generates:
- Global leaderboard: token usage ranking across all platforms
- 2D/3D contribution graphs: like GitHub's contribution graph, but showing your token consumption heatmap
For people juggling multiple AI coding tools, this is a real pain point. Previously you had to manually check each platform's dashboard. Now one command pulls everything.
Installation
cargo install tokscale
It collects data by reading local logs and config files from each platform, so you don't need to actively report anything. This means relative privacy safety — data stays local.
Community signal
271 stars in a week, 2,689 total. Ranked 4th on the Rust weekly trending page. The issue section already has feature requests coming in, which means people are actually using it.
Limitations
Right now it does usage tracking and visualization, not cost optimization suggestions (like "this task would be 40% cheaper with Model B"). No automatic routing or usage alerts either. These are reasonable future directions, but for now it's purely a "see where your money goes" tool.
For individual developers, if your AI coding spend has passed the "not worth worrying about" threshold — say, above $50/month — tokscale is worth installing. Understanding where the money goes matters more than blindly trying to save.
Primary source: