CodeBurn Shows Developers Exactly Where Their AI Coding Budget Disappears
CodeBurn is a free, local-only TUI that breaks down every token and dollar you spend across Claude Code, Codex, Cursor, and more — with waste detection built in.
Takeaways
- CodeBurn is a free, open-source terminal dashboard that breaks down AI coding spend by task, model, tool, and project across 25+ tools including Claude Code, Codex, and Cursor.
- It reads session files directly from disk — no API keys, no proxies, no data leaves your machine.
- The
codeburn optimizecommand scans for waste patterns (unused MCP servers, low cache hits, bloated CLAUDE.md) and returns copy-paste fixes with dollar savings estimates. - A one-shot rate metric tracks what percentage of code edits succeeded without retry loops, surfacing where your agent is burning tokens on fix cycles.
- Install via
npm install -g codeburnor run instantly withnpx codeburn; a native macOS menubar app is also available. - The project is MIT-licensed and available at github.com/getagentseal/codeburn.
If you use Claude Code, Codex, or Cursor daily, you probably know the feeling: the bill arrives and you have no idea where the money went. You pay for Claude, Codex, Cursor, and a stack of other AI tools. The bill tells you the total. It never tells you that half of it went to conversation instead of code, or that an expensive model burned your budget on work a cheaper one would have one-shot. CodeBurn is an open-source terminal dashboard built to fix exactly that.
A dashboard that already knows everything
CodeBurn reads the session files your tools already write to disk and breaks down every token and dollar by task, model, tool, and project. Everything runs locally , no wrapper, no proxy, no API keys, nothing leaves your machine. The trick is that all the data you need has been sitting on your disk the whole time; CodeBurn is just the first tool that makes reading it effortless.
CodeBurn tracks token usage, cost, and performance across 25 AI coding tools, breaking down spending by task type, model, tool, project, and provider so you can see exactly where your budget goes. Supported providers include:
- Claude Code , reads JSONL session transcripts from
~/.claude/projects/ - Codex (OpenAI) , reads rollout JSONL files from
~/.codex/sessions/ - Cursor , reads token counts from its local SQLite database
- OpenCode , reads from SQLite databases at
~/.local/share/opencode/ - GitHub Copilot , reads session state (output tokens only)