The latest articles on DEV Community by Davi (@davi_6869ffcaf10967a32fe2). https://dev.to/davi_6869ffcaf10967a32fe2 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4002860%2F1a034574-4132-4348-9f4c-0baf949c934c.jpg https://dev.to/davi_6869ffcaf10967a32fe2 en Davi Thu, 25 Jun 2026 18:44:34 +0000 https://dev.to/davi_6869ffcaf10967a32fe2/how-i-built-a-free-tool-that-shows-where-claude-code-burns-tokens-2hj0 https://dev.to/davi_6869ffcaf10967a32fe2/how-i-built-a-free-tool-that-shows-where-claude-code-burns-tokens-2hj0 <h2> The problem </h2> <p>Just saying "hi" to Claude Code costs <strong>~31,000 tokens</strong>.</p> <p>I was paying $500+/month in API costs and had no idea where the tokens were going. So I built <strong>tokenwise</strong> — a free CLI that shows exactly where your AI coding agent wastes tokens.</p> <h2> What it does </h2> <h3> <code>tokenwise audit</code> — Scan your instruction files </h3> <p>It scans your CLAUDE.md, AGENTS.md, and rules files, then shows:</p> <ul> <li>How many tokens each file costs <strong>per message</strong> </li> <li>Boilerplate the AI already knows ("Always write clean code")</li> <li>ALL-CAPS emphasis that doesn't help (NEVER, ALWAYS, MUST)</li> <li>Duplicate sections</li> <li>Unscoped rules that load when they shouldn't</li> </ul> <h3> <code>tokenwise scan</code> — Analyze your sessions </h3> <p>It reads your latest session logs and shows:</p> <ul> <li>Token breakdown by component (system prompt, history, tool output)</li> <li>Cache hit rate</li> <li>Top 3 "token hogs" with actionable tips</li> <li>Monthly cost projection</li> </ul> <h2> The key insight </h2> <p>Most people try to <strong>compress</strong> context (which makes the AI dumber).</p> <p>tokenwise <strong>measures</strong> first, then applies safe fixes. You can't optimize what you can't measure.</p> <h2> Quick start </h2> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>npx @davizin713/tokenwise audit npx @davizin713/tokenwise scan </code></pre> </div> <p>Zero API calls. Zero LLM inference. 100% local. Free forever.<br> Works with 11 agents</p> <p>Claude Code, OpenCode, Cursor, Aider, Cline, Codex CLI, Goose, Continue.dev, Windsurf, Augment, and Kilocode.<br> Links</p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>GitHub: github.com/davi713albano-coder/tokenwise npm: npmjs.com/package/@davizin713/tokenwise </code></pre> </div> <p>Built with</p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>TypeScript / Node.js js-tiktoken for token counting sql.js for reading OpenCode sessions MIT licensed </code></pre> </div> <p>If you use Claude Code or any AI coding agent, try it and let me know what you think! ⭐</p> claudecode ai typescript opensource