CodeBurn

CodeBurn is an open-source CLI tool that gives developers a clear view of where their AI coding tokens and money go. It reads session data directly from disk — no wrapper, no proxy, no API keys required — and presents an interactive terminal dashboard with gradient charts, responsive panels, and keyboard navigation. It supports Claude Code, Codex (OpenAI), Cursor, OpenCode, Pi, and GitHub Copilot via a provider plugin system, and fetches live pricing from LiteLLM for accurate cost calculations across all models.

• Interactive TUI Dashboard — Run codeburn to launch a full terminal UI with daily cost charts, per-project, per-model, and per-activity breakdowns, navigable with arrow keys and keyboard shortcuts.
• Multi-Provider Support — Auto-detects Claude Code, Codex, Cursor, OpenCode, Pi, and GitHub Copilot session data on disk; toggle between providers with the --provider flag or press p in the dashboard.
• 13-Category Task Classifier — Deterministically classifies every AI turn into categories like Coding, Debugging, Feature Dev, Refactoring, Testing, and more — no LLM calls needed.
• One-Shot Rate Tracking — Measures the percentage of edit turns that succeeded without retries, so you can see where the AI nails it first try vs. burns tokens on edit/test/fix loops.
• Optimize Command — Run codeburn optimize to scan sessions and your ~/.claude/ setup for waste patterns (low cache hits, unused MCP servers, bloated CLAUDE.md files) and get copy-paste fixes with estimated token and dollar savings.
• CSV/JSON Export — Export full dashboard data as structured JSON or CSV with codeburn export or codeburn report --format json, pipeable to jq for custom filtering.
• macOS Menu Bar Widget — Install a SwiftBar plugin with codeburn install-menubar to see today's cost in the menu bar with a flame icon and a dropdown activity breakdown.
• Multi-Currency Support — Display costs in any of 162 ISO 4217 currencies; exchange rates are fetched from Frankfurter (European Central Bank data) and cached for 24 hours.
• Project Filtering — Use --project and --exclude flags on any command to scope results to specific projects or exclude noise.
• No API Keys Required — Works entirely by reading local session files; pricing data is auto-cached from LiteLLM with hardcoded fallbacks for Claude and GPT-5 models.