Skip to main content
SimilarExport ZIP
100%

Claude Code token usage analyzer - breaks down usage by project, session, and subagent

✓ Published0🌍 Public
CChrisb4
Last edited Apr 6, 2026
Created on Apr 8, 2026

This visualization analyzes token usage from Claude Code’s local project logs, stored as JSONL files in `~/.claude/projects/`, and breaks down consumption by project, session, and subagent. The Python script (using `json`, `pathlib`, and `collections.defaultdict`) parses assistant and user messages, tallying `input_tokens`, `cache_creation_input_tokens`, `cache_read_input_tokens`, and `output_tokens` from the `usage` field. It recursively processes subagent session files in a `subagents` subdirectory, identifying human prompts by filtering out tool results and sidechain entries. The output is a structured dataset showing total tokens per session and project, suitable for further charting.

AI-generated description

Similar vizzes

Loading thumbnail…

Anthropic OAuth CLI - Simplified Claude Code spoof demo

This demo shows a simplified Claude Code spoof that impersonates Anthropic's OAuth flow to harvest user authorization codes. The Python script defines an `AnthropicOAuth` class that generates PKCE challenges, constructs an authorization URL pointing to claude.ai, and exchanges returned codes for access and refresh tokens via the console.anthropic.com token endpoint. It uses the `requests` library for HTTP calls and stores credentials in a local JSON file with restricted permissions, mimicking legitimate CLI authentication behavior.

JJastor11
56% match
Loading thumbnail…

CLAUDE CODE CLI — COMANDOS VERIFICADOS (Jun 2026)

This visualization presents a verified cheat sheet of Claude Code CLI commands from June 2026, organized as a styled markdown table with installation, terminal, and flag sections. The example uses standard HTML and CSS for layout, with command names and flags highlighted through monospace typography and color-coded sections. It references official documentation from code.claude.com and docs.claude.com as the data source, while the rendered image (1254x1254 pixels) shows the complete command reference in a clean, scannable format for developers.

IIoTeacher
55% match
Loading thumbnail…

CLAUDE.md — Universal ASCII / Terminal Dashboard Generator: guidance file for Claude Code to create beautiful, production-quality ASCII, ANSI, Unicode, and TUI dashboards from simple user objectives. Includes design system, composition rules, primitives, anti-patterns, library selection, and domain templates.

This example showcases a system prompt for generating terminal-native dashboards, demonstrating how a single user objective like "Monitor my Kubernetes cluster" is translated into polished ASCII, ANSI, and Unicode mockups. The guidance file defines design principles prioritizing clarity, dense layout, and meaningful status visualization, with layout intelligence rules that infer domain-specific metrics. It defaults to Python with the Rich library for implementation, and references specific primitives and anti-patterns to guide code generation. The description covers both the conceptual design system and the practical code generation approach.

IIoTeacher
53% match
Loading thumbnail…

Claude Code JSON-to-TOON Hook

This visualization demonstrates a Claude Code hook that automatically converts JSON in prompts to TOON format, achieving 30-60% token savings. It shows how the `encode` function from the TOON library transforms repetitive JSON structures like `{"products": [{"sku": "A123", ...}]}` into compact row-based notation `products[2]{sku,name,price}:` followed by value lines. The hook reads prompts from standard input, identifies JSON code blocks and inline objects using regular expressions, applies heuristics to distinguish data from JavaScript code, and outputs the converted prompt. The example includes installation steps, configuration through `settings.json`, and manual testing via command-line echo.

JJastor11
52% match
Loading thumbnail…

Claude

This interactive explainer demonstrates three distinct AI prompting strategies—prompt crafting, financial stock analysis, and research-driven coding—within a single glassmorphism-styled interface. Users trigger scenarios through buttons to compare system-prompt variations and observe real-time behavioral outputs. The visualization combines custom CSS gradient overlays with frosted-glass panel effects, while JavaScript handles scenario switching and dynamic feedback logs. Code snippets are rendered as formatted markdown blocks, showing users practical prompt structures they can adapt for their own AI workflows.

JJastor11
52% match
Loading thumbnail…

Aider Models in OpenRouter

This scatter plot compares usage and cost of AI models listed on OpenRouter as of July 2025, with tokens on the x-axis and price per million tokens on the y-axis. Data is loaded from a CSV and visualized using D3 v7 with `d3.forceSimulation` to position labels, preventing overlap, and `d3.axisBottom`/`axisLeft` for axes. The chart highlights that Claude Sonnet 4 and Gemini 2.5 Pro lead in usage, DeepSeek V3 0324 offers low cost with substantial use, while Claude Opus appears too expensive for widespread adoption.

CCurran Kelleher
51% match
Loading thumbnail…

Claude.md File for Data Analytics

This example shows a markdown file that defines coding standards and architectural principles for building data analytics dashboards. It emphasizes loose coupling, high cohesion, and composability for visualization components, alongside strict Python code style rules like PEP 8 and type hints. The document outlines a development workflow that prioritizes simple, evolving designs and testing with edge cases. The file itself is a static configuration artifact, not a live visualization, and appears to be used as a project-guiding document rather than rendering data.

MMrSteve2
50% match
Loading thumbnail…

Gist 9ea11dbcc52b4bb1c048

This example shows performance analysis for a data engineering capstone project, comparing a legacy Python log parser against a proposed Scala implementation on Spark and Mesos. The visualization tracks transfer metrics from ASCP log data, specifically examining throughput and packet rates. Using d3.js for charting, it evaluates how efficiently the new streaming solution handles regex parsing of unstructured logs. The project also demonstrates integration with Redis for persistent storage and FaspStream for live data ingestion, with the D3 visualizations presenting benchmark results from test log datasets.

BBBischof
49% match