shubhransh-gupta

Toknt — AI skill for Claude Code

AI community

Tokens? Tokn't. Local-first token optimization for AI coding agents (Cursor, Claude Code, Codex).

How to install Toknt

This entry records only its repository, not the path inside it, so there is no exact command to give. Open shubhransh-gupta/toknt and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Toknt does

Tokens? Tokn't. Local-first token optimization for AI coding agents (Cursor, Claude Code, Codex). Cut redundant context. Keep the intelligence.

Alternatives in AI

Codeburn — Free, local tool to track AI coding token usage and cost across 37 tools and agents (Claude Code, Cursor, Code 9.7k ★
Jcodemunch MCP — Cut AI token costs 95%+ on code exploration 2.6k ★
Token Monitor — Local-first desktop widget for tracking token usage, costs, and limits across 32+ AI coding tools—including Cl 1.8k ★

README

TOKEN OPTIMIZATION FOR AI AGENTS

Cut the token waste.
Keep the intelligence.

**Local-first token optimization for Claude Code · Cursor · Codex · Windsurf · VS Code**

[![GitHub stars](https://img.shields.io/github/stars/shubhransh-gupta/toknt?style=flat-square&logo=github&color=22c55e)](https://github.com/shubhransh-gupta/toknt/stargazers) [![License: MIT](https://img.shields.io/badge/License-MIT-blue?style=flat-square)](LICENSE) [![CI](https://img.shields.io/github/actions/workflow/status/shubhransh-gupta/toknt/ci.yml?style=flat-square&logo=githubactions&label=CI)](https://github.com/shubhransh-gupta/toknt/actions) [![Observatory](https://img.shields.io/badge/Observatory-live-22c55e?style=flat-square)](https://shubhransh-gupta.github.io/toknt/)

[Observatory](https://shubhransh-gupta.github.io/toknt/) · [Setup guide](docs/reduce-token-cost.md) · [Getting started](docs/getting-started.md) · [Contributing](CONTRIBUTING.md)

Measure, compare, and reduce redundant context before it reaches your model. Your code never leaves your machine — no uploads, no cloud.

Before		After
38.2K tokens	−91.5%	3.2K tokens
balanced mode · tiktoken cl100k_base		measured

*Default `safe` mode: **~6.5%** on mixed sessions · Duplicate file reads: **~46%** · [Full stats ↓](#measured-accuracy)*

Why Tokn't

Problem	What Tokn't does
Agent rereads the same file repeatedly	Hash-based dedup + local recall URI
`npm test` dumps thousands of lines into context	Summarize failures; store full output in `~/.toknt/`
Directory listings send tens of thousands of paths	Tree summary + searchable local index
Identical grep/search results repeated	Content-hash caching
Dashboards that only report waste	Actually compresses what hits the model