junhoyeo/tokscale

What it solves

Tokscale provides a centralized way to monitor and analyze token consumption and costs across a wide variety of AI coding agents and IDEs. Instead of checking multiple separate dashboards or logs, users can track their input, output, cache, and reasoning tokens in one place to understand their spending and usage patterns.

How it works

Tokscale scans local data directories, databases, and transcripts from dozens of supported AI clients (such as Claude Code, Cursor, GitHub Copilot CLI, and Zed Agent). It uses a native Rust core for high-performance parallel file scanning and SIMD JSON parsing to aggregate this data. For pricing, it integrates with LiteLLM's pricing data to calculate real-time costs, with fallbacks to OpenRouter and specific model pricing.

Who it’s for

Developers who use multiple AI coding assistants and want a high-performance tool to visualize their token usage, track costs, and compare their consumption across different platforms.

Highlights

• Broad Platform Support: Tracks usage from over 40 different AI clients, including popular IDEs and CLI agents.
• Interactive TUI: A terminal user interface powered by Ratatui featuring six interactive views, including a GitHub-style contribution graph.
• High Performance: Built with a native Rust core for processing speeds up to 10x faster than standard methods.
• Detailed Metrics: Tracks specific token types, including input, output, cache read/write, and reasoning tokens.
• Web Visualization: Includes an interactive 2D and 3D contribution graph for visualizing usage over time.
• Task-Attributed Reports: Uses LLMs to summarize sessions and group them into tasks.