CodeBurn

CodeBurn is a free, open-source terminal dashboard that shows you exactly where your AI coding tokens and money are going. It reads session data directly from disk — no wrappers, no proxies, no API keys required — and breaks down your usage by task, project, model, and activity type. How CodeBurn works CodeBurn scans the local session files that AI coding tools already write to disk. It supports Claude Code, Claude Desktop, OpenAI Codex, Cursor, OpenCode, Pi, and GitHub Copilot out of the box, with a plugin system that makes adding new providers a single file. Once installed (npm install -g codeburn), just run codeburn in your terminal and you get an interactive dashboard with gradient charts, keyboard navigation, and responsive panels. The core differentiator is its 13-category task classifier. Instead of showing "you spent $X today," CodeBurn tells you that 21% of your spend went to debugging, 18% to feature development, 14% to refactoring, and so on — all classified deterministically with zero LLM calls. It also tracks one-shot success rate per activity type, so you can see where edit-test-fix retry loops are burning tokens. Who should use CodeBurn CodeBurn is built for developers and teams using AI coding agents who want real visibility into their costs. If you are on a Claude Code Max plan and want to understand why you are hitting limits, or if your team uses Cursor across multiple projects and needs per-project cost tracking, CodeBurn gives you that breakdown. Engineering managers who need to justify AI tool spend or identify waste patterns across a team will also find it valuable. Key features - 13-category task classifier: Coding, Debugging, Feature Dev, Refactoring, Testing, Exploration, Planning, Delegation, Git Ops, Build/Deploy, Brainstorming, Conversation, General - One-shot success rate tracking per activity to spot retry loops - Breakdowns by daily cost, project, model (Opus, Sonnet, Haiku, GPT-5, GPT-4o, Gemini), activity, core tools, shell commands, and MCP servers - codeburn optimize: scans your sessions and setup for waste patterns, returns copy-paste fixes, and grades your setup health A through F - codeburn status: compact one-liner for today and month totals - codeburn export: CSV and JSON multi-period export - Date range filtering with --from and --to flags - Project and provider filtering with --project, --exclude, and --provider flags - 162 currency support via Frankfurter exchange rate API - Native macOS menubar app with live cost tracking - JSON output mode (--format json) for piping to jq or custom scripts - Auto-refresh mode (--refresh 60) for continuous monitoring Pricing CodeBurn is free and open source under the MIT license. Install it with npm install -g codeburn or run it without installing via npx codeburn. There are no paid tiers, no API keys, and no accounts required. Why CodeBurn stands out Unlike cost tracking tools that only show per-day or per-model totals (like ccusage), CodeBurn operates at the task level. Its deterministic classifier means no extra LLM calls or latency. The optimize command goes beyond reporting — it identifies specific waste patterns like re-read files, low read-to-edit ratios, unused MCP servers, ghost agents, bloated CLAUDE.md files, and cache overhead, then gives you copy-paste fixes. It also earned 109 points on Hacker News and 2,700 GitHub stars within days of launch, reflecting strong demand for this kind of visibility.